L1 regression loss for proposals going to 0

potterhsu / easy-faster-rcnn.pytorch

An easy implementation of Faster R-CNN (https://arxiv.org/pdf/1506.01497.pdf) in PyTorch.

MIT License

165 stars 57 forks source link

L1 regression loss for proposals going to 0 #12

Open peiherng97 opened 5 years ago

peiherng97 commented 5 years ago

Hi, I noticed that the smooth L1 regression loss for the fast r cnn goes to 0 when there are no proposals that have an iou of > 0.5. Is this deliberately done? The fg_indices are empty when there are no matches

https://github.com/potterhsu/easy-faster-rcnn.pytorch/blob/2c30c6d4ea57402c813294a499181b6ad710f858/model.py#L124-L129