BranchyYOLO

A Deep Learning model built using the same idea as YOLO, but with a 2-branch topology. This is its PyTorch implementation for Risiko! table game's pieces object detection.

In this repository there is also an ablated version of YOLOv9-C which has been studied during the development of BranchyYOLO.

For more information read the report.

Requirements

• PyTorch
• YOLOv9-C official implementation: git clone 'https://github.com/WongKinYiu/yolov9.git' After downloading it run the following commands:
  • sed -i 's/opt.min_items/min_items/' yolov9/val.py
  • sed -i 's/opt.min_items/min_items/' yolov9/val_dual.py
• Install the yolov9 requirements: pip install -r yolov9/requirements.txt

Files explanation

We don't include our training dataset because it is too big

• The file run.py contains the code to train the model and also to test it

  Important: modify run.py imports according to the model being trained: use *_dual files if and only if the ablated model is going to be used (since it has the DualDetect block instead of the Detect block at the end).

• The file BranchyYOLO.yaml contains the definition of BranchyYOLO model; it will be imported by models.yolo.parse_model
• The file AblatedYOLOv9-C.yaml contains the definition of the ablated version of YOLOv9-C
• The file hyp.yaml contains the definition of some hyperparameters used during the training phase
• The file coco.yaml contains the definition of the dataset used for training, validation and testing
• The file Detection.ipynb can be used to perform object detection in images