DEV Community: ashish

building an image dataset

ashish — Sat, 21 Nov 2020 09:25:47 +0000

It's bit of hectic process in creating image datasets. It Basically consists of below mentioned pipeline.(to my understanding)

Model Bias.
Whats your model goal?
Ways to collect images.
Cleaning the data.
Resizing the images.

Model Bias

Can you solve this riddle??

A man and his son are in a terrible accident and are rushed to the hospital in critical care. The doctor looks at the boy and exclaims "I can't operate on this boy, he's my son!" How could this be?

Firstly most people generally think what i think😃, this is an example for human bias.

If you train your model with more cat images and expect it to perform well on detecting cats and dogs, this happens

source: Sidney Harris

For more details on data bias you can go through this excellent slides by cs224n: Bias in the Vision and Language of Artificial Intelligence

Ways to collect data

here's a just a sample list of sources to collect images data

Search engines 🔍 (Google, Bing, Yandex, Duck Duck Go)
Social Media > through hashtags#️⃣
Youtube videos and flickr📹
take a camera/mobile and go around collect data by yourself.

Cleaning the data.

Trash the Images which can't be loaded/ corrupted.
find out duplicate images(due to various search engines).
Do what's necessary...

Resizing the images

Resize maintaining its aspect ratio.
If you have images of different sizes, and you try using resize with padding(filling the pixels with black/white).
Smaller your images >>> faster your model training.

Codes you need(💪 open source)

Some of these requires chromedriver and selenium.

For images downloading based on Search engines:

Image Downloader by sczhengyabin [google | bing]
google images download by hardikvasa
yandex images download by bobokvsky
Bulk Bing Image downloader by ostrolucky
Flickr image-scraping software developed by Ultralytics LLC

For downloading from instagram based on hashtags:

Instagram-scraper by arc298

For duplicate images cleaning:

Imagededup by idealo✨ 😎

This "imagededup" package uses Convolutional Neural Network (CNN) and hashing algorithms to find duplicates in images.

Explore Computer Vision Datasets

ashish — Mon, 16 Nov 2020 11:39:32 +0000

This article list outs resources to find the trending data-sets in computer vision field useful for beginners and other practitioners.

Datasets

Resources

Kaggle

Google Data Search

visualdata.io

CVonline: Image Databases

YACVID

Computer Vision Online

Datasetlist

Using telegram bot for Image Classification

ashish — Mon, 16 Nov 2020 11:37:27 +0000

training the model:

Pick any of your comfortable framework PyTorch / Tensorflow. And then train your image classifier /or/ use any trained model.

Training example: tensorflow | pytorch

I'm using tflite model as it serves best for edge and low computing devices.

model conversion:

If you want to convert from pytorch to tflite you can go through this code from omerferhatt

It converts from torch to ONNX and then TF2 >>> tflite

Torch->ONNX->TF2->TFLite

Put the Inference code for this model in a '.py' file and place it along with model file and labels.txt (for tensorflow).

telegram bot:

easy steps:

1) connect with bot father

2) create your bot and get token

I used python-telegram-bot library for building the bot. Just used the simple echo-bot template example and changed according to my need.

import logging
import os

from telegram import Update
from telegram.ext import Updater, CommandHandler, MessageHandler, Filters, CallbackContext
from model import get_predictions  # calling model func

# Enable logging
logging.basicConfig(
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', level=logging.INFO
)

logger = logging.getLogger(__name__)


# Define a few command handlers. These usually take the two arguments update and
# context. Error handlers also receive the raised TelegramError object in error.
def start(update: Update, context: CallbackContext) -> None:
    """Send a message when the command /start is issued."""
    update.message.reply_text('Hi send an image to classify!')


def help_command(update: Update, context: CallbackContext) -> None:
    """Send a message when the command /help is issued."""
    update.message.reply_text('Help!')


def photo(update: Update, context: CallbackContext) -> int:
    user = update.message.from_user
    photo_file = update.message.photo[-1].get_file()
    photo_file.download('user_photo.jpg')
    logger.info("Photo of %s: %s", user.first_name, 'user_photo.jpg')
    update.message.reply_text(
        'Okay now wait a few seconds!!!'
    )
    update.message.reply_text(get_prediction('user_photo.jpg'))


def main():
    """Start the bot."""
    # Create the Updater and pass it your bot's token.
    TOKEN = " " # place your token here
    updater = Updater(TOKEN, use_context=True)
    PORT = int(os.environ.get('PORT', '8443'))

    # Get the dispatcher to register handlers
    dispatcher = updater.dispatcher

    # on different commands - answer in Telegram
    dispatcher.add_handler(CommandHandler("start", start))
    dispatcher.add_handler(CommandHandler("help", help_command))

    # on noncommand i.e message - echo the message on Telegram
    dispatcher.add_handler(MessageHandler(Filters.photo & ~Filters.command, photo))

    updater.start_webhook(listen="0.0.0.0",
                      port=PORT,
                      url_path=TOKEN)
    updater.bot.set_webhook("https://yourapp.herokuapp.com/" + TOKEN)
    updater.idle()

if __name__ == '__main__':
    main()

In above code, update.message.reply_text(get_prediction('user_photo.jpg')) I'm passing the result predictions to user by calling the model, you can replace it according to your inferencing model code.

Final touch

I choose Heroku platform for deploying my bot. For more hosting ideas visit this link

Before moving forward, you need to create requirements file and Procfile.

For tflite inference you don't need whole tensorflow library, just install the tflite interpretor. See this guide for full info.

Put these lines in Procfile according to your need:

web: python bot.py

Now my folder structure looks like this:

└── img_classify_bot
    ├── model.py
    └── bot.py
    └── model.tflite
    └── labels.txt
    └── bot.py
    └── requirements.txt
    └── Procfile

finally, use heroku-cli for deploying the bot.(Your bot Token is sensitive info, don't reveal it publicly)

classifier-bot