search

HelloRicky123 / Siamese-RPN

Full reimplementation of siamese rpn, has 0.24 eao on vot2017.
MIT License
223 stars 44 forks source link

fix former 3 conv layer #6

Closed charlesyann closed 5 years ago

charlesyann commented 5 years ago

hello, Two questions for training:

  1. Did you freeze the last convolution layer too in training phase? keys_former3conv = ['featureExtract.0.weight', 'featureExtract.0.bias', 'featureExtract.1.weight', 'featureExtract.1.bias', 'featureExtract.1.running_mean', 'featureExtract.1.running_var', 'featureExtract.4.weight', 'featureExtract.4.bias', 'featureExtract.5.weight', 'featureExtract.5.bias', 'featureExtract.5.running_mean', 'featureExtract.5.running_var', 'featureExtract.8.weight', 'featureExtract.8.bias', 'featureExtract.9.weight', 'featureExtract.9.bias', 'featureExtract.9.running_mean', 'featureExtract.9.running_var']

  2. What's the meaning of set track_running_stats to false if you have set required_grad to false? 'model.featureExtract[layer].track_running_stats = False',

HelloRicky123 commented 5 years ago

2, it will make the running mean and var fixed.

charlesyann commented 5 years ago

2, it will make the running mean and var fixed.

but as far as i know, set required_grad to false will also make running mean and var fixed, yes?

HelloRicky123 commented 5 years ago

Check the doc of pytorch or some blog