JiahuiYu / slimmable_networks

Slimmable Networks, AutoSlim, and Beyond, ICLR 2019, and ICCV 2019
Other
914 stars 131 forks source link

Size mismatch for bn running_mean and running_var, us_mobilenet_v1 #47

Open octavianmm opened 4 years ago

octavianmm commented 4 years ago

I trained the us_mobilenet_v1 on a different dataset and when loading the best_model_bn_calibrated.pt, I get lots of size mismatch errors for bn running_mean and running_var. Any idea on how to overcome these errors? Also, in your opinion is is possible to train the universally slimmable model to lower widths than 0.25 (e.g. 0.1, 0.15) ?