search

zh320 / realtime-semantic-segmentation-pytorch

PyTorch implementation of over 30 realtime semantic segmentations models, e.g. BiSeNetv1, BiSeNetv2, CGNet, ContextNet, DABNet, DDRNet, EDANet, ENet, ERFNet, ESPNet, ESPNetv2, FastSCNN, ICNet, LEDNet, LinkNet, PP-LiteSeg, SegNet, ShelfNet, STDC, SwiftNet, and support knowledge distillation, distributed training etc.
Apache License 2.0
107 stars 19 forks source link

About STDC #11

Closed chris2000520 closed 4 months ago

chris2000520 commented 4 months ago

Have you used the pre-trained weights from the STDC official repository? If used, will mIoU be improved? Looking forward to your reply！

zh320 commented 4 months ago

Hi Chris,

The models in this repo are built from scratch followed only the guidances of their corresponding papers. Only those models whose backbones are ResNet or MobileNetV2 are using pretrained weights from torchvision. So it's likely we are unable to load the pretrained weights from official repo. However, it's possible to get better results if we pretrain the model using larger dataset. As you may see, the mIoUs of some models in this repo are lower than the official implementation, which may be due to pretraining.

Best.