[GitHub] zhreshold commented on issue #10638: [Feature Request] Gluon model zoo allow fine-tuning

[GitBox <gi...@apache.org>]

zhreshold commented on issue #10638: [Feature Request] Gluon model zoo allow fine-tuning
URL: https://github.com/apache/incubator-mxnet/issues/10638#issuecomment-391819304
 
 
   This could be long term transition because training models on different dataset is not purely an engineering work.
   
   Without breaking current API, I would propose to provide the default as Imagenet 1000 class, and allow user to specify the targeting task such as 
   ```
   vision.resnet18_v1(pretrained=True)   # for the default imagnet 1000
   vision.resnet18_v1(pretrained=True, class=10)  # raise error
   vision.resnet18_v1(pretrained=True, class=10, finetune=True)   # okay, we strip off the dense layer for user implicitly and replace a new output layer
   
   
   # the above all assume pertained weights are from ImageNet
   # later we can have
   vision.resnet18_v1(pretrained=True, class=xxx, dataset=yyy) 
   ```

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services