make torch.no_grad effective

xiaoyufenfei / Efficient-Segmentation-Networks

Lightweight models for real-time semantic segmentationon PyTorch (include SQNet, LinkNet, SegNet, UNet, ENet, ERFNet, EDANet, ESPNet, ESPNetv2, LEDNet, ESNet, FSSNet, CGNet, DABNet, Fast-SCNN, ContextNet, FPENet, etc.)

MIT License

914 stars 164 forks source link

make torch.no_grad effective #5

Open dcrmg opened 4 years ago

dcrmg commented 4 years ago

make torch.no_grad effective, avoid out of memory, reset batch_size of validating