/ Games

Game Detection API using Tensorflow and Go

Game Detection API using Tensorflow and Go

gamedetect

gamedetect is a simple API that uses a trained neural network to identify games that are within the top 100 currently on Twitch (as of March 2019). The full list of supported games can be seen here. The network is trained using retrain.py which uses InceptionV3 as a pre-trained network. Honestly I'm still at the point where I have no idea what the hell I'm talking about so please bear with me.

Try this...

Start a container with the gamedetect API running by running

docker run -e DEMO=true -p 8080:8080 s32x/gamedetect

(Navigating to http://localhost:8080 will provide a visual demonstration)

Then, send a POST request with a (relatively clear) game screenshot (one in the supported list) in the "image" field of a form to localhost:8080.

curl -X POST http://localhost:8080 -F [email protected]_GAME_SCREENSHOT.png

Excited yet? I sure am! gamedetect is a fun project I've been playing with in my free time to learn about Computer Vision, Neural Networks, and Tensorflow. It's sort of my own hello world app that also could potentially serve a real use-case on Twitch or any other streaming platform that requires broadcasters to categorize their stream. That being said, I'm still very much a beginner to all of this and I'm sure I'm doing a number of things wrong - feel free to let me know in the issues if you'd like.

NOTE: The public API is currently just for demonstration purposes. It's recommended to utilize the below Docker image if you're interested in higher performance on your own machine/s.

Running with Docker

To start using gamedetect via Docker, install Docker and run docker run:

docker run -p 8080:8080 s32x/gamedetect

This will retrieve the remote DockerHub image and start the service on port 8080.

Disclaimer

The included trained graph is constantly being tweaked and retrained with new datasets as I learn more about what works and what doesn't. Your results likely won't be perfect however I'm constantly working on improving them!

GitHub