search

irfanICMLL / structure_knowledge_distillation

The official code for the paper 'Structured Knowledge Distillation for Semantic Segmentation'. (CVPR 2019 ORAL) and extension to other tasks.
BSD 2-Clause "Simplified" License
708 stars 103 forks source link

Pretrained student? #58

Closed mbahri closed 3 years ago

mbahri commented 3 years ago

The training script specifies

--student-pretrain-model-imgnet ./dataset/resnet18-imagenet.pth \

Would it be possible to get the weights please?

I've tried loading the pre-trained ResNet18 from torchvision but it didn't work because some kernels are 7x7 instead of 3x3.

I also tried loading http://sceneparsing.csail.mit.edu/model/pretrained_resnet/resnet18-imagenet.pth, loading seems to work but then the dataloader threads crash with a segmentation fault.

I can train with a student from scratch, so it seems the crash is related to loading the pre-trained model.

Any help would be greatly appreciated.

irfanICMLL commented 3 years ago

I use this pre-trained weight.

https://cloudstor.aarnet.edu.au/plus/s/fNk3BUezIlJv0Y2

BBBBchan commented 3 years ago

I use this pre-trained weight.

https://cloudstor.aarnet.edu.au/plus/s/fNk3BUezIlJv0Y2

This pre-trained weight you provide can not fit the student model in your code (I have checked out to d1ec858 branch). I follow the #44 and use this pre-trained weight. The final model achieves 0.715 MIOU at 39697 steps. Is this because I use the different pre-trained student model? If so, how can I fit this pre-trained weight into the code?