Build YOLOv1 from scratch (with only 1 anchor for simplicity, and no classification)
- Implement Non-Maximum Suppression (NMS) to filter bounding boxes.
- Build a CNN backbone (e.g., a light ResNet), a YOLO head, and handle the dataset formatting for training.
- Write a function to transform YOLO head outputs and plot the results.
- Train it on a custom dataset such as detecting planes (you can find datasets on Kaggle).