search

YuwenXiong / py-R-FCN

R-FCN with joint training and python support
MIT License
1.05k stars 471 forks source link

Training and Testing time issue #70

Open Xiangyu-CAS opened 7 years ago

Xiangyu-CAS commented 7 years ago

Hi~ I had trained and test py-R-FCN on GPU K80, before that I also did the same process with py-Faster-RCNN. However , the training and testing time of py-Faster-RCNN-VGG16 is almost close to the time mentioned in paper (train time 1.2s , test time 0.3s ), while py-R-FCN is much slower than py-Faster-RCNN (train time 2s test time 0.4s). It's far away from the time you mentioned in Readme (train time 0.13s test time 0.09s). I used joint training without OHEM (experiments/scripts/rfcn_end2end.sh) and the pretrain model is ResNet-101. Could you help me find out what's wrong?

mursalal commented 7 years ago

Hello! I have 4 gpus with 4 GB. Is it enough or need at least one gpu with 6 GB?

YuwenXiong commented 7 years ago

Hi @Xiangyu-CAS and @mursalal , please try our new implementation, https://github.com/msracver/Deformable-ConvNets. It supports multi-GPU training/test and require GPU memory.

mursalal commented 7 years ago

Thanks for the response @Orpine. I will try.