search

BWGZK / CycleMix

[CVPR 2022] CycleMix: A Holistic Strategy for Medical Image Segmentation from Scribble Supervision
86 stars 15 forks source link

Smoothness weight was larger than GCO_MAX_ENERGYTERM; danger of integer overflow. #11

Closed Leelinw closed 2 months ago

Leelinw commented 2 months ago 
   Hello, I used my own dataset with the format .nii.gz, then I ran run.sh. But when training to "Epoch: [0]  [  130/51999]  eta: 14:47:55  lr: 0.000100  loss: 1.3704 (1.4059)  loss_CrossEntropy: 1.3704 (1.4059)  time: 1.0271  data: 0.0065  max mem: 17015", it stopped and "Smoothness weight was larger than GCO_MAX_ENERGYTERM; danger of integer overflow."  came out.
   The dataset used to be .svs files, is there anything wrong when I converted the format?

question