Forked From https://github.com/longcw/faster_rcnn_pytorch
Delete almost everything form original repo except roi pooling. Now this is a pure roi-pooling repo.

Getting Start

First clone this repo on local computer. git clone https://github.com/longcw/faster_rcnn_pytorch

Make

Execute make.sh.

Demo

Example code is included in demo.py.

Other

BugFixed:

• Comment following lines to enable batch_size > 1 :
  • https://github.com/longcw/faster_rcnn_pytorch/blob/0a398a2d48f94dda19f96db1fad0f5139f533002/faster_rcnn/roi_pooling/src/roi_pooling_cuda.c#L27-L30
  • https://github.com/longcw/faster_rcnn_pytorch/blob/0a398a2d48f94dda19f96db1fad0f5139f533002/faster_rcnn/roi_pooling/src/roi_pooling_cuda.c#L69-L72
• Fix a subtle bug in the cuda backwards code for roi pooling that manifests itself only when batch size is > 1. Mentioned at longcw#52 (comment)