search

Eric-mingjie / rethinking-network-pruning

Rethinking the Value of Network Pruning (Pytorch) (ICLR 2019)
MIT License
1.51k stars 293 forks source link

Custom Dataset and architecture #21

Open abhigoku10 opened 5 years ago

abhigoku10 commented 5 years ago

@liuzhuang13 @Eric-mingjie @quelleG Thanks for the sharing the wonderful work , i just have few queries .

  1. Is the source code applicable only to the imagenet dataset or can i use to other custom dataset
  2. the architecture i have is a modified version of res can i use the source code .
  3. how much performance gain you have obtained from ur exp
Eric-mingjie commented 5 years ago
  1. We provide source code for CIFAR-10 (applicable to CIFAR-100) and ImageNet. I think in general, the code for pruning does not depend on the dataset but it depends on the architecture used.
  2. For some pruning methods, you may want to read the pruning code personally. I think for unstructured weight pruning, you can directly use the source code.
  3. What do you mean by performance gain? Our work is focused on showing the training the pruned models from scratch can match the accuracy of fine-tuning.
abhigoku10 commented 5 years ago

@Eric-mingjie thanks for the response . for the Pt 3 i have used Network slimming for pruning my resenet 18 architecture , i have modified my source code but when i train , i am not getting any predictions . I am having near to mobile net architecture , i want to perform optimization on this network to increase the fps/ prediction time . which method would be good to try it first