T-rex-detr

In this project, I fine tune DETR Model to detect T-Rex, Cactus and Bird in the T-REX Game and automatically play the game according to their positions.

This project is inspired by https://github.com/S-Nuttapong/ObjectDetection-Play-TRex-Runner

Requirements

• Python 3
• PyTorch (used version ==1.9)
• Torchvision (for transforms)
• OpenCV (used version ==4.5.3)
• PyAutoGUI
• MSS (to grab screens)

How to use ?

1. Download the model from this drive
2. To test the model on a folder of images:

python test.py --data_path [folder_path] --resume [checkpoint_path] 

3. To let the model play the game for you:

• Open the game in a browser tab
• Shrink the browser tab and put it on the top left corner # you can change the desired position by fixing the resize function in detr/Image_grab.py
• uncomment the section: (from test.py main)

while True:
       image_np, (width, height) = resize(capture_screenshot(x=0, y=0, w=200, h=200), percent=0.8)
       infer_image(image_np, model, postprocessors, device, args.output_dir)
       if cv2.waitKey() == ord('a'):
           break

• run the command:

python test.py --data_path [folder_path] --resume [checkpoint_path]