Hello everyone! I wanted to share a new project I’ve started after my bootcamp. My project wasn’t selected during the course, but I’m excited to continue developing it on my own. The idea is simple but challenging: you upload a picture of a game, and the system instantly tells you the game’s name and details.
The Challenge
I took this picture last week at Book-Off. As you can see, video game covers aren’t always easy to recognize. Titles are often in Japanese, stickers can cover parts of the artwork, and sometimes the Japanese and Western covers look completely different.
My solution is to create an interface where you upload any game image, and the system identifies the game and shows detailed information about it immediately.
From Image to Game Title
Here’s the basic flow:
Upload: The user uploads a game image.
Preprocessing: The system preprocesses the image, cropping or rotating it as needed.
Matching: The image is compared against a database of about 20,000 known game covers.
Identification: The closest match is selected, and the game title is identified.
Fetching Details: Additional information about the game is pulled from a database.
How I’m Building It
The project has four main stages:
Scraping
I collect game cover images from screenscraper.fr and game info from IGDB.com. Storing images locally allows faster training and matching.Preprocessing
Using OpenCV, I resize, crop, and correct images. Standardizing the images is crucial to ensure accurate recognition.Training
I plan to use a Convolutional Neural Network (CNN) built with TensorFlow and Keras to train the model to recognize the game covers.Deployment
Finally, I’ll deploy a web app with Streamlit. Users will be able to upload an image and get instant results.
Current Progress
I’ve already started preprocessing a database of 230,000 rows of game metadata, which will serve as the foundation for training and matching.
I’m excited to see how this project develops and will share updates as I make progress!
Top comments (0)