AtomicHack

0. Create Docker env

docker build -t welder:1.0 .
docker run -d -p 9090:80 --name welder_yolo welder:1.0 

Train model and make a submission

1. Make sure you have a GPU machine, otherwise it will take more then eternity.
2. Put your files in the YOLO-friendly format (each image supported with the .txt file with bounding boxes, more Here
3. Launch blocks in HERE, this will produce the model.
4. Use the model in a Telegram bot (TBD) or make a submission via THIS

Launch BOT

Current bot is https://t.me/welder_wizard_bot You can direct him a document, and he will send you back the one with detected flaws. How to launch it:

1. Export to enviromental vars yout telegram_bot_token and path to model as

export TG_TOKEN="MY_LOVELY_TOKEN"
export MODEL_PATH="MY_LOVELY_MODEL_PATH"

2. Launch bot.py
3. You are free to go! (Unfortunately, the bot part is not tested properely, and still can fail from time to time, but it generally works.)