<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: KAUSHIKEE KARMAKAR</title>
    <description>The latest articles on DEV Community by KAUSHIKEE KARMAKAR (@piks_dev).</description>
    <link>https://dev.to/piks_dev</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4154346%2F8af6ab7a-cde5-422b-8561-15c05cca02dd.png</url>
      <title>DEV Community: KAUSHIKEE KARMAKAR</title>
      <link>https://dev.to/piks_dev</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/piks_dev"/>
    <language>en</language>
    <item>
      <title>Spooo! The Mosquito</title>
      <dc:creator>KAUSHIKEE KARMAKAR</dc:creator>
      <pubDate>Sat, 31 Oct 2026 16:45:32 +0000</pubDate>
      <link>https://dev.to/piks_dev/spooo-the-mosquito-4jf4</link>
      <guid>https://dev.to/piks_dev/spooo-the-mosquito-4jf4</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Backyard Sweep&lt;/strong&gt; is an AI-powered mosquito breeding site detection application that helps uncover potential breeding spots hiding in ordinary household surroundings.&lt;/p&gt;

&lt;p&gt;For this week's &lt;em&gt;Ghost in the Machine&lt;/em&gt; theme, I wanted to explore a slightly different kind of invisible presence.&lt;/p&gt;

&lt;p&gt;Not every threat announces itself. Sometimes, it hides in an old tire, a forgotten bottle, a flower vase, a coconut shell, or a blocked drain. Containers that seem harmless can quietly collect standing water and become potential mosquito breeding sites.&lt;/p&gt;

&lt;p&gt;We often stop noticing these things because they're part of our everyday surroundings.&lt;/p&gt;

&lt;p&gt;That's where Backyard Sweep comes in.&lt;/p&gt;

&lt;p&gt;It turns a routine household inspection into a simple, AI-assisted walk. Users photograph areas around their homes, and an open-weight YOLOv8 detector identifies potential breeding containers, highlights them with bounding boxes, and provides straightforward action tips:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Tip it out.&lt;/li&gt;
&lt;li&gt;Scrub it.&lt;/li&gt;
&lt;li&gt;Bin it.&lt;/li&gt;
&lt;li&gt;Cover it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Instead of manually examining every object and trying to remember what needs attention, users can photograph their surroundings and let computer vision help identify potential problem areas.&lt;/p&gt;

&lt;p&gt;The application also includes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Direct Camera Access:&lt;/strong&gt; Take photographs directly from a mobile device.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Confidence Slider:&lt;/strong&gt; Filter detection results without running the model again.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Copyable To-Do List:&lt;/strong&gt; Convert detected objects into actionable household tasks.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Walk Tracker:&lt;/strong&gt; Five photographs count as a completed inspection walk.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Optional Weather Card:&lt;/strong&gt; Check live weather information through Open-Meteo before heading outside.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Light and Night Forest Modes:&lt;/strong&gt; Choose between a clean light interface and a custom dark theme.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Backyard Sweep is designed for households, community volunteers, and people living in areas where dengue and mosquito-borne diseases are recurring concerns.&lt;/p&gt;

&lt;p&gt;It is free to use and is not a medical tool or disease prediction system. Its purpose is to make identifying and removing potential mosquito breeding containers easier.&lt;/p&gt;

&lt;p&gt;The inspiration behind the project is simple: sometimes the things we need to notice most are the things we have learned to overlook.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Live Application:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Note: The application runs on Render's free server, so the first load after a period of inactivity may take up to a minute.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" alt="Backyard Sweep flagging containers in a photo from the dataset's test split" width="800" height="552"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The demonstration image uses a photograph from the VisText-Mosquito test split. My own photographs were used separately for real-world field testing.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;GitHub Repository:&lt;/strong&gt; &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;https://github.com/KK-Dev0/Mosquito-Breeding-Detector&lt;/a&gt;&lt;/p&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/KK-Dev0" rel="noopener noreferrer"&gt;
        KK-Dev0
      &lt;/a&gt; / &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;
        Mosquito-Breeding-Detector
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Backyard Sweep: YOLOv8 detector for mosquito breeding sites (tires, vases, bottles, coconut shells, drain inlets), trained on VisText-Mosquito.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Backyard Sweep&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;em&gt;Turn the weekly "check for standing water" chore into a short, guided walk.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466"&gt;&lt;img src="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466" alt="License"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162"&gt;&lt;img src="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162" alt="Python"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838"&gt;&lt;img src="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364"&gt;&lt;img src="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364" alt="YOLOv8n"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532"&gt;&lt;img src="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532" alt="Hacktoberfest"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Live demo:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="nofollow noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;
(Free server: the first load after a quiet spell can take up to a minute.)&lt;/p&gt;
&lt;p&gt;Photograph anything around your home that could hold water. A YOLOv8 detector flags the containers that commonly breed mosquitoes, and each one comes with a plain-language tip for what to do about it.&lt;/p&gt;
&lt;p&gt;Built for the Hacktoberfest 2026 Open-Source AI Challenge, Week 1: &lt;strong&gt;Touch Grass&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector/demo.png"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FKK-Dev0%2FMosquito-Breeding-Detector%2FHEAD%2Fdemo.png" alt="Backyard Sweep scanning a dataset test photo"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Example result on a photo from the VisText-Mosquito test split (not one of my own photos). Photo: Faiyaz and Jisan, CC BY 4.0.&lt;/em&gt;&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Contents&lt;/h2&gt;
&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-this-exists" rel="noopener noreferrer"&gt;Why this exists&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#features" rel="noopener noreferrer"&gt;Features&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#how-it-works" rel="noopener noreferrer"&gt;How it works&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#model-and-results" rel="noopener noreferrer"&gt;Model and results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#api" rel="noopener noreferrer"&gt;API&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-open-source-mattered" rel="noopener noreferrer"&gt;Why open source mattered&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#run-it-locally" rel="noopener noreferrer"&gt;Run it locally&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#deploy-on-render" rel="noopener noreferrer"&gt;Deploy on Render&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#project-structure" rel="noopener noreferrer"&gt;Project structure&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#limits-and-responsible-use" rel="noopener noreferrer"&gt;Limits and responsible use&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#ideas-for-next-steps" rel="noopener noreferrer"&gt;Ideas for next steps&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#credits-and-licence" rel="noopener noreferrer"&gt;Credits and licence&lt;/a&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why this exists&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Dengue is spread by &lt;em&gt;Aedes&lt;/em&gt; mosquitoes, which breed in small…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Backyard Sweep combines open-source computer vision with a lightweight web application.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. AI Model — YOLOv8n
&lt;/h3&gt;

&lt;p&gt;I used Ultralytics YOLOv8n, an open-weight object detection model, fine-tuned for 30 epochs at 640 px using a free Google Colab T4 GPU.&lt;/p&gt;

&lt;p&gt;The model was trained to recognize five types of potential mosquito breeding containers:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Bottles&lt;/li&gt;
&lt;li&gt;Coconut shells&lt;/li&gt;
&lt;li&gt;Drain inlets&lt;/li&gt;
&lt;li&gt;Tires&lt;/li&gt;
&lt;li&gt;Vases&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. Dataset
&lt;/h3&gt;

&lt;p&gt;I used the Breeding Place Detection subset of VisText-Mosquito.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;3,862 training images&lt;/li&gt;
&lt;li&gt;371 validation images&lt;/li&gt;
&lt;li&gt;183 test images&lt;/li&gt;
&lt;li&gt;Five object classes&lt;/li&gt;
&lt;li&gt;CC BY 4.0 license&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Dataset citation: Faiyaz, Md Faiyaz Abdullah Sayeedi; Jisan, Md Adnanul (2025), &lt;em&gt;VisText-Mosquito: A Multimodal Dataset for Mosquito Breeding Site Detection, Surface Segmentation, and Reasoning&lt;/em&gt;, Mendeley Data, V3, DOI: 10.17632/rtsfh7jh7p.3.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. Application Architecture
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Backend:&lt;/strong&gt; FastAPI serves the trained detection model.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend:&lt;/strong&gt; HTML, CSS, and JavaScript, without a build step.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hosting:&lt;/strong&gt; Render free web service.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Weather Integration:&lt;/strong&gt; Open-Meteo API, requested only when the user opens the optional weather card.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Image Processing:&lt;/strong&gt; Photographs are processed in memory and are not saved.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  4. Model Evaluation
&lt;/h3&gt;

&lt;p&gt;On the held-out dataset test split (183 images, 424 objects), the 30-epoch model achieved:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Result&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;mAP50&lt;/td&gt;
&lt;td&gt;0.926&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Precision&lt;/td&gt;
&lt;td&gt;0.906&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recall&lt;/td&gt;
&lt;td&gt;0.844&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Class-wise AP50:&lt;/strong&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Class&lt;/th&gt;
&lt;th&gt;Test Instances&lt;/th&gt;
&lt;th&gt;AP50&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Bottle&lt;/td&gt;
&lt;td&gt;50&lt;/td&gt;
&lt;td&gt;0.884&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Coconut shell&lt;/td&gt;
&lt;td&gt;101&lt;/td&gt;
&lt;td&gt;0.973&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain inlet&lt;/td&gt;
&lt;td&gt;54&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire&lt;/td&gt;
&lt;td&gt;115&lt;/td&gt;
&lt;td&gt;0.893&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vase&lt;/td&gt;
&lt;td&gt;104&lt;/td&gt;
&lt;td&gt;0.923&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;I also experimented with a longer 80-epoch training run, which stopped early at epoch 78. It did not outperform the original model: mAP50 decreased from 0.926 to 0.914, while recall decreased from 0.844 to 0.832.&lt;/p&gt;

&lt;p&gt;I retained the original 30-epoch model.&lt;/p&gt;

&lt;h3&gt;
  
  
  5. Taking the Model Outside the Dataset
&lt;/h3&gt;

&lt;p&gt;A model performing well on a dataset does not automatically mean it will perform equally well in unfamiliar environments.&lt;/p&gt;

&lt;p&gt;I tested Backyard Sweep using five photographs taken around my own home. This was a small exploratory test, not a formal benchmark.&lt;/p&gt;

&lt;p&gt;The results revealed several limitations:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A pile containing two clear tires and one partially hidden tire produced only one correct tire detection at 41%. A parked car was incorrectly classified as a vase at 87%, while a dark corner was classified as a coconut shell at 70%.&lt;/li&gt;
&lt;li&gt;A blue dustbin was classified as a vase at 83%.&lt;/li&gt;
&lt;li&gt;Among approximately five paint buckets and tins, only one was detected at 25%, with three additional detections at 10%.&lt;/li&gt;
&lt;li&gt;A rooftop containing two clay water pots, a plastic bucket, and a water tank produced no detections, even at 10%.&lt;/li&gt;
&lt;li&gt;An open drain channel was also missed.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This was an important discovery. Many everyday objects capable of holding water were outside the model's original five classes. Confidence scores from unfamiliar real-world photographs also could not always be trusted.&lt;/p&gt;

&lt;p&gt;The experiment reminded me that AI can reveal things hidden from ordinary observation, but it can also overlook what is right in front of it.&lt;/p&gt;

&lt;h3&gt;
  
  
  6. Experimenting with YOLO-World
&lt;/h3&gt;

&lt;p&gt;I wanted to explore whether another open model could identify objects that my fine-tuned YOLOv8n detector missed.&lt;/p&gt;

&lt;p&gt;I experimented with YOLO-World, an open-vocabulary object detector, without additional training.&lt;/p&gt;

&lt;p&gt;I supplied custom descriptions such as clay water pot, bucket, and water tank. These descriptions were selected after examining my photographs, so this was an exploratory demonstration rather than a fair independent comparison.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Fine-tuned YOLOv8n&lt;/th&gt;
&lt;th&gt;YOLO-World&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Dataset mAP50&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;0.926&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;0.297&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire recall&lt;/td&gt;
&lt;td&gt;0.80&lt;/td&gt;
&lt;td&gt;0.017&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain-inlet AP50&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;td&gt;0.000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Loaded parameters&lt;/td&gt;
&lt;td&gt;~3 million&lt;/td&gt;
&lt;td&gt;~164 million&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Model file size&lt;/td&gt;
&lt;td&gt;~6 MB&lt;/td&gt;
&lt;td&gt;329 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;YOLO-World successfully identified objects my original model missed in the rooftop photograph, including a clay pot (89%), a bucket (77%), and a water tank.&lt;/p&gt;

&lt;p&gt;However, it also produced false detections, identifying a parked car and a bathroom basin as water tanks, and confusing a real tire with a clay water pot.&lt;/p&gt;

&lt;p&gt;I ultimately retained YOLOv8n because it performed considerably better on the dataset's trained classes and was much smaller.&lt;/p&gt;

&lt;p&gt;I also investigated deploying YOLO-World. Its 329 MB model file and approximately 1.7 GB RAM requirement during my Colab test exceeded Render's free server memory limit of 512 MB.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;My next step:&lt;/strong&gt; Use YOLO-World to generate preliminary labels for additional household containers, manually correct the bounding boxes, and fine-tune the smaller YOLOv8n model using more diverse real-world photographs.&lt;/p&gt;

&lt;h3&gt;
  
  
  Development Process
&lt;/h3&gt;

&lt;p&gt;I built this project with assistance from Claude through a chat-based workflow. Claude helped develop the FastAPI application and frontend, and drafted the README and this submission.&lt;/p&gt;

&lt;p&gt;I handled model training in Colab, tested the application using my own photographs, investigated its failures, compared alternative models, evaluated the results, and made the final decisions about which model and architecture to retain.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;For Backyard Sweep, open innovation made experimentation, customization, and accessibility possible.&lt;/p&gt;

&lt;p&gt;I never benchmarked a closed AI API against this application, so I cannot claim that open-source models outperform proprietary alternatives.&lt;/p&gt;

&lt;p&gt;However, using open-weight models gave me several important advantages.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Freedom to Fine-Tune&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I could fine-tune a relatively small, approximately 3-million-parameter detector for a specific household inspection task without paying per-image inference fees.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Freedom to Experiment&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I could switch between YOLOv8n and YOLO-World, evaluate both on the same dataset, investigate their limitations, and choose the architecture that best suited my deployment requirements.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Greater Control Over Data&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Photographs are processed on a server I control rather than being sent to a third-party AI inference service. They are processed in memory without being saved.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. Transparency and Reproducibility&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The source code, model weights, dataset reference, and licensing information are publicly accessible. Other developers can inspect the implementation, reproduce the experiments, and adapt the project.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Accessibility&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;A lightweight model can run on relatively inexpensive infrastructure, making the application accessible without expensive hardware or individual API subscriptions.&lt;/p&gt;

&lt;p&gt;There are trade-offs, too. Ultralytics YOLO uses the AGPL-3.0 license, so the project is also licensed under AGPL-3.0. The free Render server also sleeps when idle.&lt;/p&gt;

&lt;p&gt;For me, open innovation is about having the freedom to experiment, question results, understand limitations, and build something that can evolve beyond its original dataset.&lt;/p&gt;

&lt;p&gt;It also means that the technology behind the project isn't an unexplained black box. The model, its weights, and the code can be examined and improved by others.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;Halloween and the idea of ghosts often bring to mind things that exist just outside our ordinary perception.&lt;/p&gt;

&lt;p&gt;Backyard Sweep approaches that idea from a different direction.&lt;/p&gt;

&lt;p&gt;There are no supernatural entities hiding in the corners of a backyard, but there can be overlooked objects quietly collecting standing water. A discarded tire, a forgotten bucket, or an unnoticed container can become a potential breeding site.&lt;/p&gt;

&lt;p&gt;The interesting part of building this project was discovering how much a machine could identify in an ordinary photograph, and equally, how much it could still miss.&lt;/p&gt;

&lt;p&gt;When I took the application outside and tested it around my own home, the model's limitations became just as interesting as its successful detections. It reminded me that making something visible is only the beginning; understanding what the machine cannot see matters just as much.&lt;/p&gt;

&lt;p&gt;I wanted to build something that encourages people to look a little closer at the spaces they inhabit.&lt;/p&gt;

&lt;p&gt;Perhaps the most important things hiding in plain sight aren't ghosts at all.&lt;/p&gt;

&lt;p&gt;They're the things we simply stopped noticing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Render:&lt;/strong&gt; The FastAPI backend serving the AI detection model is deployed using Render's free web service.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;em&gt;Credits: Open-Meteo for providing the optional live weather information.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Mosquito Banner</title>
      <dc:creator>KAUSHIKEE KARMAKAR</dc:creator>
      <pubDate>Sat, 31 Oct 2026 16:39:06 +0000</pubDate>
      <link>https://dev.to/piks_dev/mosquito-banner-1mk9</link>
      <guid>https://dev.to/piks_dev/mosquito-banner-1mk9</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week3-2026-10-19"&gt;Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Backyard Sweep&lt;/strong&gt; is an AI-powered mosquito breeding site detection application that makes routine household inspections easier, quicker, and a little less tedious.&lt;/p&gt;

&lt;p&gt;Mosquitoes breed in standing water, often in places we stop noticing: old tires, bottles, vases, coconut shells, buckets, and blocked drains. Health agencies recommend regularly emptying, scrubbing, or covering these containers. But manually checking every corner of a home can feel like another repetitive chore.&lt;/p&gt;

&lt;p&gt;That's the problem I wanted to address.&lt;/p&gt;

&lt;p&gt;Backyard Sweep turns that weekly inspection into a short walk. Instead of manually identifying every possible breeding container, users simply photograph areas around their homes. An open-weight YOLOv8 detector identifies potential breeding spots, highlights them with bounding boxes, and provides straightforward action tips:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Tip it out.&lt;/li&gt;
&lt;li&gt;Scrub it.&lt;/li&gt;
&lt;li&gt;Bin it.&lt;/li&gt;
&lt;li&gt;Cover it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The application also includes a few features designed to reduce unnecessary effort:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Direct Camera Access:&lt;/strong&gt; Take photographs directly from a mobile device without navigating through additional steps.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Confidence Slider:&lt;/strong&gt; Filter detection results without running the model again.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Copyable To-Do List:&lt;/strong&gt; Convert detected objects into actionable household tasks.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Walk Tracker:&lt;/strong&gt; Five photographs count as a completed inspection walk.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Optional Weather Card:&lt;/strong&gt; Check live weather conditions through Open-Meteo before heading outside.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Light and Night Forest Modes:&lt;/strong&gt; Choose between a clean light interface and a custom dark theme.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The idea is simple: spend less time figuring out what needs attention and more time actually taking care of it.&lt;/p&gt;

&lt;p&gt;Backyard Sweep is designed for households, community volunteers, and people living in areas where dengue and mosquito-borne diseases are recurring concerns. It is free to use and is not a medical tool or disease prediction system. Its purpose is to make identifying and removing potential mosquito breeding containers less of a chore.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Live Application:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Note: The application runs on Render's free server, so the first load after a period of inactivity may take up to a minute.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" alt="Backyard Sweep flagging containers in a photo from the dataset's test split" width="800" height="552"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The demonstration image uses a photograph from the VisText-Mosquito test split. My own photographs were used separately for real-world field testing.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;GitHub Repository:&lt;/strong&gt; &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;https://github.com/KK-Dev0/Mosquito-Breeding-Detector&lt;/a&gt;&lt;/p&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/KK-Dev0" rel="noopener noreferrer"&gt;
        KK-Dev0
      &lt;/a&gt; / &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;
        Mosquito-Breeding-Detector
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Backyard Sweep: YOLOv8 detector for mosquito breeding sites (tires, vases, bottles, coconut shells, drain inlets), trained on VisText-Mosquito.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Backyard Sweep&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;em&gt;Turn the weekly "check for standing water" chore into a short, guided walk.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466"&gt;&lt;img src="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466" alt="License"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162"&gt;&lt;img src="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162" alt="Python"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838"&gt;&lt;img src="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364"&gt;&lt;img src="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364" alt="YOLOv8n"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532"&gt;&lt;img src="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532" alt="Hacktoberfest"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Live demo:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="nofollow noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;
(Free server: the first load after a quiet spell can take up to a minute.)&lt;/p&gt;
&lt;p&gt;Photograph anything around your home that could hold water. A YOLOv8 detector flags the containers that commonly breed mosquitoes, and each one comes with a plain-language tip for what to do about it.&lt;/p&gt;
&lt;p&gt;Built for the Hacktoberfest 2026 Open-Source AI Challenge, Week 1: &lt;strong&gt;Touch Grass&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector/demo.png"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FKK-Dev0%2FMosquito-Breeding-Detector%2FHEAD%2Fdemo.png" alt="Backyard Sweep scanning a dataset test photo"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Example result on a photo from the VisText-Mosquito test split (not one of my own photos). Photo: Faiyaz and Jisan, CC BY 4.0.&lt;/em&gt;&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Contents&lt;/h2&gt;
&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-this-exists" rel="noopener noreferrer"&gt;Why this exists&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#features" rel="noopener noreferrer"&gt;Features&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#how-it-works" rel="noopener noreferrer"&gt;How it works&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#model-and-results" rel="noopener noreferrer"&gt;Model and results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#api" rel="noopener noreferrer"&gt;API&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-open-source-mattered" rel="noopener noreferrer"&gt;Why open source mattered&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#run-it-locally" rel="noopener noreferrer"&gt;Run it locally&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#deploy-on-render" rel="noopener noreferrer"&gt;Deploy on Render&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#project-structure" rel="noopener noreferrer"&gt;Project structure&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#limits-and-responsible-use" rel="noopener noreferrer"&gt;Limits and responsible use&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#ideas-for-next-steps" rel="noopener noreferrer"&gt;Ideas for next steps&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#credits-and-licence" rel="noopener noreferrer"&gt;Credits and licence&lt;/a&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why this exists&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Dengue is spread by &lt;em&gt;Aedes&lt;/em&gt; mosquitoes, which breed in small…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Backyard Sweep combines open-source computer vision with a lightweight web application.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. AI Model — YOLOv8n
&lt;/h3&gt;

&lt;p&gt;I used Ultralytics YOLOv8n, an open-weight object detection model, fine-tuned for 30 epochs at 640 px using a free Google Colab T4 GPU.&lt;/p&gt;

&lt;p&gt;The model recognizes five categories of potential mosquito breeding containers:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Bottles&lt;/li&gt;
&lt;li&gt;Coconut shells&lt;/li&gt;
&lt;li&gt;Drain inlets&lt;/li&gt;
&lt;li&gt;Tires&lt;/li&gt;
&lt;li&gt;Vases&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. Dataset
&lt;/h3&gt;

&lt;p&gt;I trained the model using the Breeding Place Detection subset of VisText-Mosquito.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;3,862 training images&lt;/li&gt;
&lt;li&gt;371 validation images&lt;/li&gt;
&lt;li&gt;183 test images&lt;/li&gt;
&lt;li&gt;Five object classes&lt;/li&gt;
&lt;li&gt;CC BY 4.0 license&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Dataset citation: Faiyaz, Md Faiyaz Abdullah Sayeedi; Jisan, Md Adnanul (2025), &lt;em&gt;VisText-Mosquito: A Multimodal Dataset for Mosquito Breeding Site Detection, Surface Segmentation, and Reasoning&lt;/em&gt;, Mendeley Data, V3, DOI: 10.17632/rtsfh7jh7p.3.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. Application Architecture
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Backend:&lt;/strong&gt; FastAPI serves the trained detection model.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend:&lt;/strong&gt; HTML, CSS, and JavaScript, with no build step.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hosting:&lt;/strong&gt; Render free web service.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Weather Integration:&lt;/strong&gt; Open-Meteo API, requested only when the user opens the optional weather card.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Image Processing:&lt;/strong&gt; Photographs are processed in memory and are not saved.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  4. Model Evaluation
&lt;/h3&gt;

&lt;p&gt;On the held-out dataset test split (183 images, 424 objects), the 30-epoch model achieved:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Result&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;mAP50&lt;/td&gt;
&lt;td&gt;0.926&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Precision&lt;/td&gt;
&lt;td&gt;0.906&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recall&lt;/td&gt;
&lt;td&gt;0.844&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Class-wise AP50:&lt;/strong&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Class&lt;/th&gt;
&lt;th&gt;Test Instances&lt;/th&gt;
&lt;th&gt;AP50&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Bottle&lt;/td&gt;
&lt;td&gt;50&lt;/td&gt;
&lt;td&gt;0.884&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Coconut shell&lt;/td&gt;
&lt;td&gt;101&lt;/td&gt;
&lt;td&gt;0.973&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain inlet&lt;/td&gt;
&lt;td&gt;54&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire&lt;/td&gt;
&lt;td&gt;115&lt;/td&gt;
&lt;td&gt;0.893&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vase&lt;/td&gt;
&lt;td&gt;104&lt;/td&gt;
&lt;td&gt;0.923&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;I also experimented with a longer 80-epoch training run, which stopped early at epoch 78. It did not outperform the original model: mAP50 decreased from 0.926 to 0.914, while recall decreased from 0.844 to 0.832.&lt;/p&gt;

&lt;p&gt;I retained the original 30-epoch model.&lt;/p&gt;

&lt;h3&gt;
  
  
  5. Real-World Testing
&lt;/h3&gt;

&lt;p&gt;A model performing well on a dataset does not automatically mean it will perform equally well outside the training environment.&lt;/p&gt;

&lt;p&gt;I tested Backyard Sweep using five photographs taken around my own home. This was a small exploratory test, not a formal benchmark.&lt;/p&gt;

&lt;p&gt;The results revealed some interesting limitations:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A pile containing two clear tires and one partially hidden tire produced only one correct tire detection at 41%. A parked car was incorrectly classified as a vase at 87%, while a dark corner was classified as a coconut shell at 70%.&lt;/li&gt;
&lt;li&gt;A blue dustbin was classified as a vase at 83%.&lt;/li&gt;
&lt;li&gt;Among approximately five paint buckets and tins, only one was detected at 25%, with three additional detections at 10%.&lt;/li&gt;
&lt;li&gt;A rooftop containing two clay water pots, a plastic bucket, and a water tank produced no detections, even at 10%.&lt;/li&gt;
&lt;li&gt;An open drain channel was also missed.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This showed that many everyday household containers were outside the model's original five classes. It also highlighted that confidence scores from unfamiliar real-world photographs cannot always be trusted.&lt;/p&gt;

&lt;h3&gt;
  
  
  6. Experimenting with YOLO-World
&lt;/h3&gt;

&lt;p&gt;I wanted to explore whether another open model could identify objects that my fine-tuned YOLOv8n detector missed.&lt;/p&gt;

&lt;p&gt;I experimented with YOLO-World, an open-vocabulary object detector, without additional training. I supplied custom descriptions such as clay water pot, bucket, and water tank.&lt;/p&gt;

&lt;p&gt;These descriptions were selected after examining my photographs, so this was an exploratory demonstration rather than a fair independent comparison.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Fine-tuned YOLOv8n&lt;/th&gt;
&lt;th&gt;YOLO-World&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Dataset mAP50&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;0.926&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;0.297&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire recall&lt;/td&gt;
&lt;td&gt;0.80&lt;/td&gt;
&lt;td&gt;0.017&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain-inlet AP50&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;td&gt;0.000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Loaded parameters&lt;/td&gt;
&lt;td&gt;~3 million&lt;/td&gt;
&lt;td&gt;~164 million&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Model file size&lt;/td&gt;
&lt;td&gt;~6 MB&lt;/td&gt;
&lt;td&gt;329 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;YOLO-World successfully identified objects my original model missed in the rooftop photograph, including a clay pot (89%), a bucket (77%), and a water tank.&lt;/p&gt;

&lt;p&gt;However, it also produced false detections, identifying a parked car and a bathroom basin as water tanks, and confusing a real tire with a clay water pot.&lt;/p&gt;

&lt;p&gt;I ultimately retained YOLOv8n because it performed considerably better on the dataset's trained classes and was much smaller.&lt;/p&gt;

&lt;p&gt;I also investigated deploying YOLO-World. Its 329 MB model file and approximately 1.7 GB RAM requirement during my Colab test exceeded Render's free server memory limit of 512 MB.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;My next step:&lt;/strong&gt; Use YOLO-World to generate preliminary labels for additional household containers, manually correct the bounding boxes, and fine-tune the smaller YOLOv8n model using more diverse real-world photographs.&lt;/p&gt;

&lt;h3&gt;
  
  
  Development Process
&lt;/h3&gt;

&lt;p&gt;I built this project with assistance from Claude through a chat-based workflow. Claude helped develop the FastAPI application and frontend, and drafted the README and this submission.&lt;/p&gt;

&lt;p&gt;I handled model training in Colab, tested the application using my own photographs, investigated its failures, compared alternative models, evaluated the results, and made the final decisions about which model and architecture to retain.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;For Backyard Sweep, open innovation made experimentation, customization, and accessibility possible.&lt;/p&gt;

&lt;p&gt;I never benchmarked a closed AI API against this application, so I cannot claim that open-source models outperform proprietary alternatives. However, using open-weight models gave me several important advantages.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Less Dependency on Paid AI Services&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I could fine-tune a relatively small, approximately 3-million-parameter detector for a specific household inspection task without paying per-image inference fees.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Freedom to Experiment&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I could switch between YOLOv8n and YOLO-World, evaluate both on the same dataset, investigate their limitations, and choose the architecture that best suited my deployment requirements.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Greater Control Over Data&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Photographs are processed on a server I control rather than being sent to a third-party AI inference service. They are processed in memory without being saved.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. Transparency and Reproducibility&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The source code, model weights, dataset reference, and licensing information are publicly accessible. Other developers can inspect the implementation, reproduce the experiments, and adapt the project.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Accessible Deployment&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;A lightweight model can run on relatively inexpensive infrastructure, making the application accessible without expensive hardware or individual API subscriptions.&lt;/p&gt;

&lt;p&gt;There are trade-offs, too. Ultralytics YOLO uses the AGPL-3.0 license, so the project is also licensed under AGPL-3.0. The free Render server also sleeps when idle.&lt;/p&gt;

&lt;p&gt;For me, open innovation is about having the freedom to experiment, understand limitations, improve existing solutions, and build something that can evolve over time.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Lazy Payoff
&lt;/h2&gt;

&lt;p&gt;The biggest effort-saving idea behind Backyard Sweep is that users don't have to manually identify and remember every potential breeding container they encounter during an inspection.&lt;/p&gt;

&lt;p&gt;Instead of:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Walking around the house.&lt;/li&gt;
&lt;li&gt;Inspecting every container individually.&lt;/li&gt;
&lt;li&gt;Trying to remember which objects need attention.&lt;/li&gt;
&lt;li&gt;Making a mental or written checklist.&lt;/li&gt;
&lt;li&gt;Repeating the process during the next inspection.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Users can photograph their surroundings, review the AI-generated detections, and use the copyable to-do list to organize the actions they need to take.&lt;/p&gt;

&lt;p&gt;The confidence slider also avoids unnecessary model re-scanning, while the five-photo tracker gives users a simple way to monitor their inspection activity.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;But how much time does it actually save?&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I haven't conducted a timed comparison between manual inspections and AI-assisted inspections yet, so I won't invent a time-saving percentage.&lt;/p&gt;

&lt;p&gt;I did test the application on five photographs from my own home. That testing also revealed that the detector needs improvement when dealing with unfamiliar household objects and environments.&lt;/p&gt;

&lt;p&gt;The goal isn't to eliminate the actual work of emptying, scrubbing, or covering containers. Those actions still require human effort.&lt;/p&gt;

&lt;p&gt;It's to make the process of noticing, identifying, and organizing that work a little easier.&lt;/p&gt;

&lt;p&gt;Sometimes being lazy isn't about avoiding responsibility. It's about finding a smarter way to get started.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Render:&lt;/strong&gt; The FastAPI backend serving the AI detection model is deployed using Render's free web service.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;em&gt;Credits: Open-Meteo for providing the optional live weather information.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Backyard Sweep: a ten-minute walk to find mosquito breeding spots, and what my own photos</title>
      <dc:creator>KAUSHIKEE KARMAKAR</dc:creator>
      <pubDate>Sat, 31 Oct 2026 16:34:37 +0000</pubDate>
      <link>https://dev.to/piks_dev/backyard-sweep-a-ten-minute-walk-to-find-mosquito-breeding-spots-and-what-my-own-photos-2656</link>
      <guid>https://dev.to/piks_dev/backyard-sweep-a-ten-minute-walk-to-find-mosquito-breeding-spots-and-what-my-own-photos-2656</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12"&gt;Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Backyard Sweep&lt;/strong&gt; is an AI-powered mosquito breeding site detection tool that turns routine household inspections into a simple, guided activity.&lt;/p&gt;

&lt;p&gt;Mosquitoes breed in standing water, often hiding in things we overlook: old tires, bottles, vases, coconut shells, buckets, and blocked drains. Health agencies recommend regularly emptying, scrubbing, or covering these containers. But the real challenge is noticing them in the first place.&lt;/p&gt;

&lt;p&gt;Backyard Sweep makes that process easier. Users can photograph areas around their homes, and an open-weight YOLOv8 detector identifies potential mosquito breeding containers, highlights them with bounding boxes, and provides straightforward action tips: tip it out, scrub it, bin it, or cover it.&lt;/p&gt;

&lt;p&gt;The application also includes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Direct Camera Access:&lt;/strong&gt; Take photographs directly from a mobile device.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Confidence Slider:&lt;/strong&gt; Filter detection results without running the model again.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Copyable To-Do List:&lt;/strong&gt; Turn detected objects into actionable household tasks.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Walk Tracker:&lt;/strong&gt; Five photographs count as a completed inspection walk.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Weather Card:&lt;/strong&gt; Optional live weather information through Open-Meteo to help decide whether it's a good time for an outdoor inspection.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Light and Dark Modes:&lt;/strong&gt; Including a custom Night Forest theme.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  The Side Hustle Potential
&lt;/h3&gt;

&lt;p&gt;Backyard Sweep is designed for households, community volunteers, housing societies, and local awareness initiatives in areas where dengue and mosquito-borne diseases are a recurring concern.&lt;/p&gt;

&lt;p&gt;It could evolve into a lightweight household inspection service or a community-focused mosquito prevention platform. Its open-source foundation also makes it possible to adapt the detector to different household environments and regional container types.&lt;/p&gt;

&lt;p&gt;The application is currently free to use. It is not a medical tool and does not diagnose or predict diseases. Its purpose is to make identifying and removing potential mosquito breeding containers easier.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Live Application:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Note: The application runs on Render's free server, so the first load after a period of inactivity may take up to a minute.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" alt="Backyard Sweep flagging containers in a photo from the dataset's test split" width="800" height="552"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The demonstration image uses a photograph from the VisText-Mosquito test split. My own photographs were used separately for real-world field testing.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;GitHub Repository:&lt;/strong&gt; &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;https://github.com/KK-Dev0/Mosquito-Breeding-Detector&lt;/a&gt;&lt;/p&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/KK-Dev0" rel="noopener noreferrer"&gt;
        KK-Dev0
      &lt;/a&gt; / &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;
        Mosquito-Breeding-Detector
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Backyard Sweep: YOLOv8 detector for mosquito breeding sites (tires, vases, bottles, coconut shells, drain inlets), trained on VisText-Mosquito.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Backyard Sweep&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;em&gt;Turn the weekly "check for standing water" chore into a short, guided walk.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466"&gt;&lt;img src="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466" alt="License"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162"&gt;&lt;img src="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162" alt="Python"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838"&gt;&lt;img src="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364"&gt;&lt;img src="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364" alt="YOLOv8n"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532"&gt;&lt;img src="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532" alt="Hacktoberfest"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Live demo:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="nofollow noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;
(Free server: the first load after a quiet spell can take up to a minute.)&lt;/p&gt;
&lt;p&gt;Photograph anything around your home that could hold water. A YOLOv8 detector flags the containers that commonly breed mosquitoes, and each one comes with a plain-language tip for what to do about it.&lt;/p&gt;
&lt;p&gt;Built for the Hacktoberfest 2026 Open-Source AI Challenge, Week 1: &lt;strong&gt;Touch Grass&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector/demo.png"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FKK-Dev0%2FMosquito-Breeding-Detector%2FHEAD%2Fdemo.png" alt="Backyard Sweep scanning a dataset test photo"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Example result on a photo from the VisText-Mosquito test split (not one of my own photos). Photo: Faiyaz and Jisan, CC BY 4.0.&lt;/em&gt;&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Contents&lt;/h2&gt;
&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-this-exists" rel="noopener noreferrer"&gt;Why this exists&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#features" rel="noopener noreferrer"&gt;Features&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#how-it-works" rel="noopener noreferrer"&gt;How it works&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#model-and-results" rel="noopener noreferrer"&gt;Model and results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#api" rel="noopener noreferrer"&gt;API&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-open-source-mattered" rel="noopener noreferrer"&gt;Why open source mattered&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#run-it-locally" rel="noopener noreferrer"&gt;Run it locally&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#deploy-on-render" rel="noopener noreferrer"&gt;Deploy on Render&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#project-structure" rel="noopener noreferrer"&gt;Project structure&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#limits-and-responsible-use" rel="noopener noreferrer"&gt;Limits and responsible use&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#ideas-for-next-steps" rel="noopener noreferrer"&gt;Ideas for next steps&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#credits-and-licence" rel="noopener noreferrer"&gt;Credits and licence&lt;/a&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why this exists&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Dengue is spread by &lt;em&gt;Aedes&lt;/em&gt; mosquitoes, which breed in small…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;The project combines an open-weight computer vision model with a lightweight web application.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. AI Model — YOLOv8n&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I used Ultralytics YOLOv8n, fine-tuned for 30 epochs at 640 px using a free Google Colab T4 GPU.&lt;/p&gt;

&lt;p&gt;The model was trained to recognize five types of potential mosquito breeding containers:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Bottles&lt;/li&gt;
&lt;li&gt;Coconut shells&lt;/li&gt;
&lt;li&gt;Drain inlets&lt;/li&gt;
&lt;li&gt;Tires&lt;/li&gt;
&lt;li&gt;Vases&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;2. Dataset&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I used the Breeding Place Detection subset of VisText-Mosquito:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;3,862 training images&lt;/li&gt;
&lt;li&gt;371 validation images&lt;/li&gt;
&lt;li&gt;183 test images&lt;/li&gt;
&lt;li&gt;Five object classes&lt;/li&gt;
&lt;li&gt;CC BY 4.0 license&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Dataset citation: Faiyaz, Md Faiyaz Abdullah Sayeedi; Jisan, Md Adnanul (2025), &lt;em&gt;VisText-Mosquito: A Multimodal Dataset for Mosquito Breeding Site Detection, Surface Segmentation, and Reasoning&lt;/em&gt;, Mendeley Data, V3, DOI: 10.17632/rtsfh7jh7p.3.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Application Architecture&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Backend:&lt;/strong&gt; FastAPI serves the trained detection model.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend:&lt;/strong&gt; HTML, CSS, and JavaScript, without a build step.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hosting:&lt;/strong&gt; Render free web service.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Weather Integration:&lt;/strong&gt; Open-Meteo API, requested only when the user opens the optional weather card.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Image Processing:&lt;/strong&gt; Photographs are processed in memory and are not saved.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Model Evaluation
&lt;/h3&gt;

&lt;p&gt;On the held-out dataset test split (183 images, 424 objects), the 30-epoch model achieved:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Result&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;mAP50&lt;/td&gt;
&lt;td&gt;0.926&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Precision&lt;/td&gt;
&lt;td&gt;0.906&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recall&lt;/td&gt;
&lt;td&gt;0.844&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Class-wise AP50:&lt;/strong&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Class&lt;/th&gt;
&lt;th&gt;Test Instances&lt;/th&gt;
&lt;th&gt;AP50&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Bottle&lt;/td&gt;
&lt;td&gt;50&lt;/td&gt;
&lt;td&gt;0.884&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Coconut shell&lt;/td&gt;
&lt;td&gt;101&lt;/td&gt;
&lt;td&gt;0.973&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain inlet&lt;/td&gt;
&lt;td&gt;54&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire&lt;/td&gt;
&lt;td&gt;115&lt;/td&gt;
&lt;td&gt;0.893&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vase&lt;/td&gt;
&lt;td&gt;104&lt;/td&gt;
&lt;td&gt;0.923&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;I also experimented with a longer 80-epoch training run, which stopped early at epoch 78. Surprisingly, it did not outperform the original model: mAP50 decreased from 0.926 to 0.914, while recall decreased from 0.844 to 0.832. I retained the original 30-epoch model.&lt;/p&gt;

&lt;h3&gt;
  
  
  Real-World Testing: Taking It Outside
&lt;/h3&gt;

&lt;p&gt;Dataset performance is one thing. Real-world performance is another.&lt;/p&gt;

&lt;p&gt;I tested Backyard Sweep using five photographs taken around my own home. This was a small exploratory test, not a formal benchmark.&lt;/p&gt;

&lt;p&gt;The results exposed several limitations:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A pile containing two clear tires and one partially hidden tire produced only one correct tire detection at 41%. Meanwhile, a parked car was incorrectly classified as a vase at 87%, and a dark corner as a coconut shell at 70%.&lt;/li&gt;
&lt;li&gt;A blue dustbin was classified as a vase at 83%.&lt;/li&gt;
&lt;li&gt;Among approximately five paint buckets and tins, only one was detected at 25%, with three additional detections at 10%.&lt;/li&gt;
&lt;li&gt;A rooftop containing two clay water pots, a plastic bucket, and a water tank produced no detections, even at 10%.&lt;/li&gt;
&lt;li&gt;An open drain channel was also missed.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This showed me that many household objects capable of holding water were outside the model's original five classes. It also demonstrated that confidence scores from unfamiliar real-world photographs cannot always be trusted.&lt;/p&gt;

&lt;h3&gt;
  
  
  Experimenting with YOLO-World
&lt;/h3&gt;

&lt;p&gt;To explore a possible improvement, I tested YOLO-World, an open-vocabulary object detector, without additional training.&lt;/p&gt;

&lt;p&gt;I supplied a custom list of object descriptions, including clay water pots, buckets, and water tanks. These descriptions were chosen after examining my own photographs, so this was an exploratory demonstration rather than a fair independent comparison.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Metric&lt;/th&gt;
&lt;th&gt;Fine-tuned YOLOv8n&lt;/th&gt;
&lt;th&gt;YOLO-World&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Dataset mAP50&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;0.926&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;0.297&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire recall&lt;/td&gt;
&lt;td&gt;0.80&lt;/td&gt;
&lt;td&gt;0.017&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain-inlet AP50&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;td&gt;0.000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Loaded parameters&lt;/td&gt;
&lt;td&gt;~3 million&lt;/td&gt;
&lt;td&gt;~164 million&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Model file size&lt;/td&gt;
&lt;td&gt;~6 MB&lt;/td&gt;
&lt;td&gt;329 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;YOLO-World successfully identified objects my original model missed in the rooftop photograph, including a clay pot (89%), a bucket (77%), and a water tank.&lt;/p&gt;

&lt;p&gt;However, it also produced false detections, identifying a parked car and a bathroom basin as water tanks, and even confusing a real tire with a clay water pot.&lt;/p&gt;

&lt;p&gt;Its dataset evaluation also used the original class names, meaning alternative text descriptions could change the results.&lt;/p&gt;

&lt;p&gt;I ultimately retained YOLOv8n because it performed considerably better on its trained dataset classes and was much smaller.&lt;/p&gt;

&lt;p&gt;I also investigated deploying YOLO-World. Its 329 MB model file and approximately 1.7 GB RAM requirement during my Colab test exceeded the free Render server's 512 MB memory limit.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;My next step:&lt;/strong&gt; Use YOLO-World to generate preliminary labels for additional household containers, manually correct the bounding boxes, and fine-tune the smaller YOLOv8n model using more diverse real-world photographs.&lt;/p&gt;

&lt;h3&gt;
  
  
  Development Process
&lt;/h3&gt;

&lt;p&gt;I built this project with assistance from Claude through a chat-based workflow. Claude helped develop the FastAPI application and frontend, and drafted the README and this submission.&lt;/p&gt;

&lt;p&gt;I handled the model training in Colab, tested the deployed application using my own photographs, investigated its failures, compared alternative models, evaluated the results, and made the final decisions about which model and architecture to retain.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;I never benchmarked a closed AI API against Backyard Sweep, so I cannot claim that open-source models outperform proprietary alternatives.&lt;/p&gt;

&lt;p&gt;However, open innovation made several things possible:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. Custom Model Training&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I could fine-tune a relatively small, approximately 3-million-parameter detector for a specific household inspection task without paying per-image API fees.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Freedom to Experiment&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I could switch between YOLOv8n and YOLO-World, evaluate both against the same dataset, investigate their limitations, and decide which was more suitable for deployment.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. Greater Control Over Data&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Photographs are processed on a server I control rather than being sent to a third-party AI inference service. The application processes them in memory without saving them.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. Transparency and Reproducibility&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The source code, model weights, dataset reference, and licensing information are publicly accessible. Anyone interested can inspect the implementation, reproduce the experiments, or adapt the project.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Accessibility&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;A lightweight model can run on a relatively inexpensive server, making the application accessible without requiring expensive infrastructure or individual API subscriptions.&lt;/p&gt;

&lt;p&gt;There are trade-offs, too. Ultralytics YOLO uses the AGPL-3.0 license, so the project is also licensed under AGPL-3.0. The free Render server also sleeps when idle.&lt;/p&gt;

&lt;p&gt;For me, open innovation is not simply about using freely available models. It is about having the freedom to experiment, question results, understand limitations, and build something that can evolve beyond its original dataset.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Render:&lt;/strong&gt; The FastAPI backend serving the AI detection model is deployed using Render's free web service.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;em&gt;Credits: Open-Meteo for providing the optional live weather information.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Backyard Sweep: a ten-minute walk to find mosquito breeding spots, and what my own photos taught me</title>
      <dc:creator>KAUSHIKEE KARMAKAR</dc:creator>
      <pubDate>Sat, 10 Oct 2026 14:25:31 +0000</pubDate>
      <link>https://dev.to/piks_dev/backyard-sweep-a-ten-minute-walk-to-find-mosquito-breeding-spots-and-what-my-own-photos-taught-me-371a</link>
      <guid>https://dev.to/piks_dev/backyard-sweep-a-ten-minute-walk-to-find-mosquito-breeding-spots-and-what-my-own-photos-taught-me-371a</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05"&gt;Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Mosquitoes breed in standing water, and the water usually sits in things we stop seeing: an old tire, a vase, a bottle, a coconut shell, a blocked drain. Health agencies advise emptying, scrubbing or covering them regularly. The hard part isn't knowing what to do. It's noticing the containers.&lt;/p&gt;

&lt;p&gt;Backyard Sweep turns that weekly check into a short walk. You go outside, photograph anything that could hold water, and an open-weight YOLOv8 detector draws a flag on each likely breeding spot, with a plain-language tip: tip it out, scrub it, bin it, cover it. On a phone, "Take a photo" opens the camera directly, so the screen is the shortest part of the experience.&lt;/p&gt;

&lt;p&gt;A few extras: a confidence slider that filters results without re-scanning, a "copy to-do list" button, and a small tracker where five photos count as a proper walk. There's also an optional "is it a good time for a lap?" card. It uses live weather from Open-Meteo, only when you tap it, and it is plain rules, not AI. There's a light mode and a "night forest" dark mode.&lt;/p&gt;

&lt;p&gt;It's for anyone who lives where dengue season is a worry and wants a free, quick way to check their own home. It is not a medical tool, and I make no health claims. Emptying containers is what reduces mosquitoes. This just makes that check feel like less of a chore.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;Live app (free server, so the first load after a quiet spell can take up to a minute): &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fds67eykmvcbdty28zhq6.png" alt="Backyard Sweep flagging containers in a photo from the dataset's test split" width="800" height="552"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;A photo from the VisText-Mosquito test split, not one of my own. My own photos are in the field test below.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/KK-Dev0" rel="noopener noreferrer"&gt;
        KK-Dev0
      &lt;/a&gt; / &lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;
        Mosquito-Breeding-Detector
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Backyard Sweep: YOLOv8 detector for mosquito breeding sites (tires, vases, bottles, coconut shells, drain inlets), trained on VisText-Mosquito.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Backyard Sweep&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;em&gt;Turn the weekly "check for standing water" chore into a short, guided walk.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466"&gt;&lt;img src="https://camo.githubusercontent.com/eb80ae5bc09563456cf2ebd5703d4be9c9b72726abd3e5a517d5ce65d6c64ae6/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6c6963656e73652d4147504c2d2d332e302d326537643466" alt="License"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162"&gt;&lt;img src="https://camo.githubusercontent.com/eb968e355f577598c22b8573c8838bf2e0454aa0d6f02022de4df7b8b7523b9d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312d333737366162" alt="Python"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838"&gt;&lt;img src="https://camo.githubusercontent.com/e4bef6d1aed3ef11e6c35e38e8b4c35bdce8252393acee088d1769392be0865a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d303039363838" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364"&gt;&lt;img src="https://camo.githubusercontent.com/9b4819b0aa87b8be0f42308c63135b1f47559f95c72338005b4598e46a176adb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f6d6f64656c2d594f4c4f76386e2d666663383364" alt="YOLOv8n"&gt;&lt;/a&gt;
&lt;a rel="noopener noreferrer nofollow" href="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532"&gt;&lt;img src="https://camo.githubusercontent.com/c767bc3ef3b17fe3fcc6fd294eb343d4d8138909942a466042157757233aad4a/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362d386132626532" alt="Hacktoberfest"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Live demo:&lt;/strong&gt; &lt;a href="https://mosquito-breeding-detector.onrender.com" rel="nofollow noopener noreferrer"&gt;https://mosquito-breeding-detector.onrender.com&lt;/a&gt;
(Free server: the first load after a quiet spell can take up to a minute.)&lt;/p&gt;
&lt;p&gt;Photograph anything around your home that could hold water. A YOLOv8 detector flags the containers that commonly breed mosquitoes, and each one comes with a plain-language tip for what to do about it.&lt;/p&gt;
&lt;p&gt;Built for the Hacktoberfest 2026 Open-Source AI Challenge, Week 1: &lt;strong&gt;Touch Grass&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector/demo.png"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FKK-Dev0%2FMosquito-Breeding-Detector%2FHEAD%2Fdemo.png" alt="Backyard Sweep scanning a dataset test photo"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;em&gt;Example result on a photo from the VisText-Mosquito test split (not one of my own photos). Photo: Faiyaz and Jisan, CC BY 4.0.&lt;/em&gt;&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Contents&lt;/h2&gt;
&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-this-exists" rel="noopener noreferrer"&gt;Why this exists&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#features" rel="noopener noreferrer"&gt;Features&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#how-it-works" rel="noopener noreferrer"&gt;How it works&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#model-and-results" rel="noopener noreferrer"&gt;Model and results&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#api" rel="noopener noreferrer"&gt;API&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#why-open-source-mattered" rel="noopener noreferrer"&gt;Why open source mattered&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#run-it-locally" rel="noopener noreferrer"&gt;Run it locally&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#deploy-on-render" rel="noopener noreferrer"&gt;Deploy on Render&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#project-structure" rel="noopener noreferrer"&gt;Project structure&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#limits-and-responsible-use" rel="noopener noreferrer"&gt;Limits and responsible use&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#ideas-for-next-steps" rel="noopener noreferrer"&gt;Ideas for next steps&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector#credits-and-licence" rel="noopener noreferrer"&gt;Credits and licence&lt;/a&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Why this exists&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Dengue is spread by &lt;em&gt;Aedes&lt;/em&gt; mosquitoes, which breed in small…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/KK-Dev0/Mosquito-Breeding-Detector" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Model:&lt;/strong&gt; YOLOv8n (open weights, Ultralytics), fine-tuned for 30 epochs at 640 px on a free Colab T4.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Data:&lt;/strong&gt; the Breeding Place Detection part of VisText-Mosquito (3,862 training images, 371 validation, 183 test, five classes: bottle, coconut shell, drain inlet, tire, vase), CC BY 4.0. Faiyaz, Md Faiyaz Abdullah Sayeedi; Jisan, Md Adnanul (2025), &lt;em&gt;VisText-Mosquito: A Multimodal Dataset for Mosquito Breeding Site Detection, Surface Segmentation, and Reasoning&lt;/em&gt;, Mendeley Data, V3, DOI 10.17632/rtsfh7jh7p.3&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;App:&lt;/strong&gt; FastAPI serves the model and one plain HTML/CSS/JS page with no build step. It's hosted on a free Render web service. Photos are processed in memory and not saved.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;Results on the dataset's held-out test split (183 images, 424 objects):&lt;/strong&gt; mAP50 0.926, precision 0.906, recall 0.844.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Class&lt;/th&gt;
&lt;th&gt;Test instances&lt;/th&gt;
&lt;th&gt;AP50&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Bottle&lt;/td&gt;
&lt;td&gt;50&lt;/td&gt;
&lt;td&gt;0.884&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Coconut shell&lt;/td&gt;
&lt;td&gt;101&lt;/td&gt;
&lt;td&gt;0.973&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain inlet&lt;/td&gt;
&lt;td&gt;54&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire&lt;/td&gt;
&lt;td&gt;115&lt;/td&gt;
&lt;td&gt;0.893&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Vase&lt;/td&gt;
&lt;td&gt;104&lt;/td&gt;
&lt;td&gt;0.923&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;I also trained a longer run (80 epochs, stopped early at epoch 78). It did not beat the 30-epoch model on the test split (mAP50 0.914 vs 0.926, recall 0.832 vs 0.844), so I kept the original.&lt;/p&gt;

&lt;p&gt;That looks great, but the test photos come from the same collection as the training photos. So I took it outside.&lt;/p&gt;

&lt;h2&gt;
  
  
  I Took It Outside, and It Struggled
&lt;/h2&gt;

&lt;p&gt;I scanned 5 of my own photos from around my home at the default 25% setting. This is a small sample, a sanity check and not a benchmark.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A pile with 2 clear tires and 1 half-hidden:&lt;/strong&gt; it found one tire, at 41%. In the same photo, a parked car was labelled "vase" at 87%, and a dark corner "coconut shell" at 70%. Both false alarms scored higher than the real tire, so no slider setting fixes it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A blue dustbin:&lt;/strong&gt; labelled "vase" at 83%. The label is wrong, but it is a real water-holding container, so the tip still fits.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;About five paint buckets and tins:&lt;/strong&gt; one flagged at 25%, three at 10%, all called vases.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A rooftop corner with two clay water pots, a plastic bucket and a water tank:&lt;/strong&gt; nothing, even at 10%.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;An open drain channel:&lt;/strong&gt; nothing.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Most of what holds water around my home isn't in the model's five classes, and on photos unlike its training data, its confidence scores can't be trusted as a filter.&lt;/p&gt;

&lt;h2&gt;
  
  
  Trying a Different Open Model
&lt;/h2&gt;

&lt;p&gt;Since open models are swappable, I tried YOLO-World, an open-vocabulary detector, with no training and a word list I typed in ("clay water pot", "bucket", "water tank" and so on). I chose those words after seeing my photos, so treat this as a demo, not a fair test.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;My fine-tuned YOLOv8n&lt;/th&gt;
&lt;th&gt;YOLO-World (zero-shot)&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;mAP50 on the dataset test split&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;0.926&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;0.297&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tire recall&lt;/td&gt;
&lt;td&gt;0.80&lt;/td&gt;
&lt;td&gt;0.017&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Drain-inlet AP50&lt;/td&gt;
&lt;td&gt;0.956&lt;/td&gt;
&lt;td&gt;0.000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Parameters when loaded&lt;/td&gt;
&lt;td&gt;about 3 million&lt;/td&gt;
&lt;td&gt;about 164 million (includes its text encoder)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Saved model file&lt;/td&gt;
&lt;td&gt;about 6 MB&lt;/td&gt;
&lt;td&gt;329 MB&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;On my rooftop photo it found what my model missed: the big clay pot (89%), the bucket (77%) and the water tank. But it also called a parked car a "water tank" (62%), a bathroom basin a "water tank" (72%), and a real tire a "clay water pot" (56%). Its dataset scores used plain class names, and other wording would change them.&lt;/p&gt;

&lt;p&gt;So the two models are good at opposite things. I kept the small one, because it's far more accurate on what it knows. I also checked whether YOLO-World could run in the app: even with the word list baked in, the saved model is a 329 MB file and used about 1.7 GB of RAM in my Colab test, far over the free server's 512 MB, so I didn't deploy it. The next step I'd take is using YOLO-World to pre-label household containers, fixing the boxes by hand, and fine-tuning the small model on my own photos.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How I worked:&lt;/strong&gt; I built this with help from Claude in a chat. It helped write the FastAPI app and the web page, and drafted the README and this post. I ran the training in Colab, tested the app on my own photos, found the failures, and decided what to keep, including keeping the original model over the longer-trained one.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;I never tested a closed API, so I can't claim open beat it on accuracy. What open weights let me do:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Fine-tune a tiny model for one job.&lt;/strong&gt; A 3-million-parameter detector trained on five container types runs on a free CPU instance, with no per-photo fee.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Swap models in minutes.&lt;/strong&gt; Testing YOLO-World against my own model took one Colab cell. Because both are open, I could score them on the same test split and see that zero-shot was far worse on the trained classes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep photos on a server I control.&lt;/strong&gt; Photos of someone's home go to my own server, not a third-party AI service. You could also run it on your own laptop.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Check everything.&lt;/strong&gt; Code, weights, dataset and licenses are public, so every number above can be reproduced.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The trade-offs: Ultralytics YOLO is AGPL-3.0, so the project is AGPL-3.0 too. And the free server sleeps when idle.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;Best Use of Render: the FastAPI server that runs the model is hosted on Render's free web service.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Credits: Open-Meteo for the optional weather card.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Stuck Doctor</title>
      <dc:creator>KAUSHIKEE KARMAKAR</dc:creator>
      <pubDate>Sun, 04 Oct 2026 17:33:38 +0000</pubDate>
      <link>https://dev.to/piks_dev/stuck-doctor-362f</link>
      <guid>https://dev.to/piks_dev/stuck-doctor-362f</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-weekend-2026-10-01"&gt;Hacktoberfest Weekend Challenge: Build for a Friend&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;I built Stuck Doctor for my friend Piku, who hesitates a lot before going to the doctor whenever she faces a health issue.&lt;/p&gt;

&lt;p&gt;Watching a friend put it off made me realize that waiting isn't laziness. It's usually fear, worry about money, awkwardness, not knowing where to start, or not knowing what to say. So instead of trying to replace a doctor, Stuck Doctor figures out why someone is stuck and gives them &lt;strong&gt;one tiny first step&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;What it does:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Speak or type in English, Hindi or Bengali, with read-aloud answers&lt;/li&gt;
&lt;li&gt;Works out why you're avoiding care and gives you a 2-minute first move&lt;/li&gt;
&lt;li&gt;Lets you rehearse booking a call with Meera, an AI clinic receptionist&lt;/li&gt;
&lt;li&gt;Family and friend mode: gentle messages to encourage someone you love to go&lt;/li&gt;
&lt;li&gt;Report decoder: checks values against the ranges printed on your report&lt;/li&gt;
&lt;li&gt;Medicine schedule and reminders&lt;/li&gt;
&lt;li&gt;After-visit notes: a clean summary and to-do list from what the doctor said&lt;/li&gt;
&lt;li&gt;A "show the doctor" card for people too shy to say it out loud&lt;/li&gt;
&lt;li&gt;Free-care finder (eSanjeevani, government hospitals, Jan Aushadhi, Ayushman Bharat)&lt;/li&gt;
&lt;li&gt;An always-available SOS button (108, 112, 104, live location on WhatsApp)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It deliberately does &lt;strong&gt;not&lt;/strong&gt; diagnose, suggest medicines or doses, or read scans. It's a nudge, not a doctor.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;Live app: &lt;a href="https://kk-dev0.github.io/Stuck_Doctor/" rel="noopener noreferrer"&gt;&lt;/a&gt;&lt;br&gt;
&lt;a href="https://kk-dev0.github.io/Stuck_Doctor/" rel="noopener noreferrer"&gt;https://kk-dev0.github.io/Stuck_Doctor/&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;(The AI backend may take a short while to wake up on the first visit.)&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/KK-Dev0/Stuck_Doctor" rel="noopener noreferrer"&gt;&lt;/a&gt;&lt;br&gt;
&lt;a href="https://github.com/KK-Dev0/Stuck_Doctor" rel="noopener noreferrer"&gt;https://github.com/KK-Dev0/Stuck_Doctor&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;MIT licensed.&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;The AI engine is &lt;strong&gt;Gemma&lt;/strong&gt;, Google's open-weight model family, run with &lt;strong&gt;Ollama&lt;/strong&gt;. The backend uses &lt;strong&gt;Gradio&lt;/strong&gt; and is hosted on &lt;strong&gt;Render&lt;/strong&gt;, with a static front end on GitHub Pages. I built it in one day at Hacktoberfest 2026.&lt;/p&gt;

&lt;p&gt;Gemma handles the language-heavy parts: understanding why someone is avoiding care, writing the first-step suggestion, playing Meera in the call rehearsal, writing messages for friends and family, explaining report values, and summarizing the doctor's instructions. It also lets the app respond in English, Hindi and Bengali.&lt;/p&gt;

&lt;p&gt;I added safety limits on purpose: the app never diagnoses or recommends medicines, summaries use only what the user wrote, report values are compared with the ranges printed on the report, and there's an optional "double-check" mode that reads a photo a second time. Emergency help is always one tap away. AI assistance was used in building this project, which is also disclosed on the site.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;A health tool for people who are already worried about money and access shouldn't depend on a paid API that could change its price or rules.&lt;/p&gt;

&lt;p&gt;An open-weight model made it possible to build something this wide in a single day, support Indian languages, and keep the code open under the MIT license so others can fork it, translate it or improve its safety. It fits the theme perfectly: AI belongs to everyone, including the friend who has been too nervous, too busy or too broke to see a doctor.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Gemma&lt;/li&gt;
&lt;li&gt;Ollama&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Single Participation
&lt;/h2&gt;

&lt;p&gt;-Kaushikee Karmakar &lt;a href="https://dev.to/piks_dev"&gt;&lt;/a&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>weekendchallenge</category>
      <category>hf26challenge</category>
    </item>
  </channel>
</rss>
