qingzhouzhen opened a new issue #7820: Low accuracy of pvanet
URL: https://github.com/apache/incubator-mxnet/issues/7820
 
 
   I have changed a model which called pvanet from caffe 
[https://github.com/sanghoon/pva-faster-rcnn](https://github.com/sanghoon/pva-faster-rcnn),
 it is a implementation of the article by the author 
[https://arxiv.org/abs/1608.08021](https://arxiv.org/abs/1608.08021), my 
problem is why my accuracy is low than author's
   What I have done:
   1 Implement the classification net: 
[https://github.com/qingzhouzhen/incubator-mxnet/blob/pvanet-e2e/example/image-classification/symbols/pvanet.py](https://github.com/qingzhouzhen/incubator-mxnet/blob/pvanet-e2e/example/image-classification/symbols/pvanet.py)
   2 feed the classification to Faster R-CNN module: 
[https://github.com/qingzhouzhen/incubator-mxnet/blob/pvanet-e2e/example/rcnn/rcnn/symbol/symbol_pvanet.py](https://github.com/qingzhouzhen/incubator-mxnet/blob/pvanet-e2e/example/rcnn/rcnn/symbol/symbol_pvanet.py)
   Most job I have done is to atapt Faster-R-CNN,  replace the vgg pre-trained 
model with pvanet pre-trained model, for details see #7786 
   Firstly the classification net pvanet's accuracy is 64.0%(70.6% as article)
   Secondly detection net (pvanet+rpn+rcnn)'s mAP is 59.37%(82.5% as article)
   I checked my net structure carefully to make it the same as author's 
implementation in caffe, is there anything I do wrong?  
 
----------------------------------------------------------------
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:
[email protected]


With regards,
Apache Git Services

Reply via email to