Hero Name Recognition

This repository provides a CNN baseline (backbone from timm) for Hero Name Recognition problem in a game called League of Legends: Wild Rift (a.k.a Wild Rift).
This game is an mutiplayer online battle arena game developed by Riot Games. In the highlight moment detection systems, it is important to recognize the hero appearing on the message bar when a battle happens.

For example:

Image	Hero
	Ahri
	Ashe

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Dependencies

pip3 install -r requirements.txt

Data

Put your data into the data folder. The structure of the folder will be:

./
    data/
        train/
            img1.jpg
            img2.jpg
            ...
        test/
            img1.jpg
            img2.jpg
            ...

Logging

This repository uses Comet ML for experiment tracking. See this for how to get started and log in with your account. Then, go to train.py and modify line 17, in the argument api_key for your own project.

Training

You can choose any model from timm to make the backbone. Here is resnet18:

python3 train.py --backbone resnet18 --e 100 --b 64 

Prediction

Make sure the --backbone to predict is the same as the training.

python3 predict.py --backbone resnet18 --d <path/to/test/data> --c <path/to/weight>

For quickly prediction without training, download the trained resnet18 weight here into checkpoint folder and run:

python3 predict.py --backbone resnet18 --d data/train --c checkpoint/weight_42_3.7745168209075928.pt

Use the flag -h to see other training or inference arguments.

Author

Nguyen Huy An
Email: anhuynguyen001@gmail.com