Implement Light-Head RCNN by modifying the src of RFCN

[sherryxie1]

Because the structure of the RFCN and light-head rcnn is pretty similar, I just tried to implement the light-head rcnn by modifying the py-R-FCN on caffe.

I only changed the prototxt file for training and testing, the training procedure goes well, but when I do test, I find the mAP is very low.

The attachments are the train and test file I used, I am really eager to know in which step I make mistakes. Could anyone help me to find out it?

resnet50_lhrcnn_test.txt resnet50_lhrcnn_train.txt

[kptikku]

Sherry, Where you able to solve the issue that you faced above?

[ZhiHuaYin]

@sherryxie1 have u successfully implemented the light head rcnn on caffe??I also have a very low mAP with my light head.