yzcjtr / GeoNet

Code for GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose (CVPR 2018)
MIT License
726 stars 181 forks source link

Criteria to select the best model #52

Closed jmendozais closed 5 years ago

jmendozais commented 5 years ago

I just checked the loss of GeoNet on training, and its seems that it has a high variance. How did you selected your best model for evaluation on depth and camera motion tasks?.

yzcjtr commented 5 years ago

Hi, you are right. Training on the KITTI dataset is painful for the loss of scene diversity. You can select the KITTI 2012 training set as a validation set.