search

xinntao / EDVR

Winning Solution in NTIRE19 Challenges on Video Restoration and Enhancement (CVPR19 Workshops) - Video Restoration with Enhanced Deformable Convolutional Networks. EDVR has been merged into BasicSR and this repo is a mirror of BasicSR.
https://github.com/xinntao/BasicSR
1.49k stars 318 forks source link

About the value error: empty range for randrange()(0,0,0) #94

Open Carcy opened 5 years ago

Carcy commented 5 years ago

I use this work to solve deblur-compression. In first stage, the warning : offset mean >100 also appears with woTSA.yml, and I ignore it on purpose. I stop the training process in 1.8w state. In second stage, the error in title is in 8000 iterations.

train.py, line 160, in main for _, train_data in enumerate(train_loader): /root/anaconda3/env/py37env/python3.7/site-packages/torch/utils/data/dataloader.py,line 560 , in next batch=self.collate_fn([self.dataset[i] for i in indices]) ......... /root/anaconda3/envs/py37env/lib/python3.7/random.py, line 200, in randrange raise valueerror("(empty range for randrange() (%d,%d,%d))"%(istart,istop,width))

xinntao commented 5 years ago
  1. I think the errors are from the data. Check your data is OK or not.
  2. If the offsets are large than 100, the offsets are meaningless, and the performance is usually poor. EDVR-L is fragile due to the dcn. Usually, we stop it and resume it from a normal model whose offsets are normal.