search

Eniac-Xie / faster-rcnn-resnet

ResNet Implementation for Faster-rcnn
MIT License
206 stars 117 forks source link

add resnet-50 prototxt #7

Closed lixiang-ucas closed 7 years ago

lixiang-ucas commented 7 years ago

add support for Faster RCNN+ResNet-50.

Eniac-Xie commented 7 years ago

Thank you! Have you test the net on pascal voc? Would you please upload the resnet-50 you trained and give us the link?

lixiang-ucas commented 7 years ago

I have test the net on my own dataset (containing only one class), not yet on pascal voc dataset.

bq-li commented 7 years ago

@Eniac-Xie .Thank you for your code. But I want to know how you train ResNet101_BN_SCALE_Merged.caffemodel which is used to initialize the model parameters。

646677064 commented 7 years ago

@lixiang-ucas thanks a lot. @Eniac-Xie I tried to change class-aware to class-agnotic(fg or bg, 2 class for bbox regression ) as R-FCN did,but not save much memory (class=6 ,RPN_POST_NMS_TOP_N=1000),and I trained resnet-101 on my own dataset failed. I think the readonly brach take up much menory