search

liuzechun / MetaPruning

MetaPruning: Meta Learning for Automatic Neural Network Channel Pruning. In ICCV 2019.
MIT License
351 stars 74 forks source link

why the val result of PruningNet model is 0 #34

Closed diligent321 closed 4 years ago

diligent321 commented 4 years ago

Hello, your MetaPruning performance is excellent.

However, when i start the PruningNet train, the val accuracy is always 0. my steps are as follows,

  1. Download the MetaPruning repository code, and cd mobilenetv2/training directory.

  2. Change the batch_size and learning_rate in train.py according to https://github.com/liuzechun/MetaPruning/issues/2#issuecomment-519879543. The default values are 512 and 0.25 respectively. Since i have only 4 2080Ti GPU cards on single machine, to avoid the gpu out of memory error, reduce them by half, 256 and 0.125.

  3. Split the imagenet train data to subtrain and subval5w, and set traindir and valdir in train.py to these two directory.

  4. Run the run.sh script file, the log is as picture, image

so where am i wrong, looking forward to your suggestions.

diligent321 commented 4 years ago

The same as https://github.com/liuzechun/MetaPruning/issues/4#issuecomment-521489128, closed.