search

torrvision / crfasrnn

This repository contains the source code for the semantic image segmentation method described in the ICCV 2015 paper: Conditional Random Fields as Recurrent Neural Networks. http://crfasrnn.torr.vision/
Other
1.34k stars 462 forks source link

The problem of using the latest caffe #86

Open huangdunbo opened 7 years ago

huangdunbo commented 7 years ago

I use the latest caffe with cudnn from https://github.com/bittnt/caffe/tree/crfrnn to run crfasrnn_demo.py for test.Using the image: input1 And change spatial_filter_weight and bilateral_filter_weight as spatial_filter_weights_str and bilateral_filter_weights_str in TVG_CRFRNN_new_deploy.prototxt The output is: output2 If I use old caffe without cudnn from https://github.com/bittnt/caffe/tree/70856cd28a10e2592b4d5ba1ae05cf6b59824a37,the output is: output Why use the latest caffe but effect worse?

xiewei198908 commented 7 years ago

@huangdunbo you need set crop parameter

huangdunbo commented 7 years ago

@xiewei198908 THX~ Note: CropLayer should set the offset parameter~See FCN's prototxt.

wk910930 commented 7 years ago

Here is the link to the FCN's prototxt.

pythonnewer commented 7 years ago

@huangdunbo I met same questions 。have you solved this problems,please.

chaoxinzheng commented 6 years ago

what is the crop parameter to get to the high performance? i tried axis 2 and offset 5 but no luck at all.