Trangle opened a new issue #3852: training faster-rcnn using resnet but the 
test detection seems not right
URL: https://github.com/apache/incubator-mxnet/issues/3852
 
 
   When I  tried to apply resnet(50, 101) to faster-rcnn?it works during the 
training process, but seemed to fail in the testing process.
   
   the symbol was like this:
   
   **symbol_resnet_full.py**
   
[symbol_resnet_full.txt](https://github.com/dmlc/mxnet/files/594099/symbol_resnet_full.txt)
   
   
   **symbol_resnet.py**
   
[symbol_resnet.txt](https://github.com/dmlc/mxnet/files/594100/symbol_resnet.txt)
   
   
   and I trained it using e2e, changed code `sym = 
get_resnet(num_classes=num_classes)` in line 35.
   
   
   the accuracy during training process at epoch 8 of rcnn was about 94%?and 
rpn about 98%?but the test result was really low with mean ap 0.22?even at the 
training set. I used resnet-50 pretrained model downloaded from dmlc as conv 
initial.

----------------------------------------------------------------
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:
us...@infra.apache.org


With regards,
Apache Git Services

Reply via email to