Algolzw / BSRT

Pytorch code for "BSRT: Improving Burst Super-Resolution with Swin Transformer and Flow-Guided Deformable Alignment", CVPRW, 1st place in NTIRE 2022 BurstSR Challenge (real-world track).
MIT License
179 stars 13 forks source link

About "FlowGuidedDCN: Offset mean is 276.96575927734375, larger than 100." #16

Open fanjunkai1 opened 10 months ago

fanjunkai1 commented 10 months ago

Has anyone encountered this issue before?

Algolzw commented 10 months ago

It means the learned offset is too large. One possible solution is to change the learning rate (smaller) and retrain the model.