BiSeNet based on pytorch 0.4.1 and python 3.6
epoch_295.pth in GoogleDrive and put it in
The net structure has been modified, so the pretrained model does not match the current model.
Use tensorboard to see the real-time loss and accuracy
loss on train
pixel precision on val
Cropped and Resized means two image processing method to make the input image size fixed, it seems like Cropped input images get better result. I guess it's because cropped masks keep the original ground truth information while resized loss it.
- Finish real-time segmentation with camera or pre-load video
- Update pixel accuracy calculation algorithm