CoD4 Screenshot Classifier

This repo contains training and inference code for a CNN-based CoD4 screenshot classifier to detect hackers. (Currently deployed on the NamelessNoobs servers.)

Inference Usage:

1. Clone the repository and install required libraries using pip install -r requirements.txt
2. Download the pretrained model from Google Drive and place the file in lightning_logs/version_1/checkpoints
3. Run predict_batches.py e.g. python predict_batches.py --image_dir=path/to/folder --threshold=0.1. The script takes the following arguments:

• image_dir path to the image folder
• batch_size inference batch size, adjust according to available memory. Default = 8.
• threshold probability threshold for classifying a screenshot as a hacker. Default = 0.5.

4. The script prints and returns predictions for all images, where 1 is the hacking class, and 0 is the clean class.