All other Conv2D layers have bias=False since they are followed by a batch-norm layer which essentially has a bias term. However, the final layer does not have batch-norm at the output and should have a bias term. This is how it is in the reference YOLO implementation.
All other
Conv2Dlayers have bias=False since they are followed by a batch-norm layer which essentially has a bias term. However, the final layer does not have batch-norm at the output and should have a bias term. This is how it is in the reference YOLO implementation.https://github.com/vietnguyen91/Yolo-v2-pytorch/blob/9589413b5dce0476eb9cccc41945cf30cf131b34/src/yolo_net.py#L64