Merge the fast-rcnn-cleanup branch into the official Caffe to avoid straying from the Caffe master branch and increasing the maintenance cost to keep updated

[futurely]

No description provided.

[y22ma]

+1

[hazirbas]

Hi,

Did anyone manage to merge fast-rcnn into caffe master branch?

Thanks.

[Austriker]

@futurely @y22ma @canerhazirbas @rbgirshick
I have created a PR 4163 based on the commit used in py-faster-rcnn.

It's almost ready to merge but I have two issues with the test that I haven't solved yet !
If you have any ideas !

[zimenglan-sysu-512]

@Austriker i merge the caffe-fast-rcnn with the [newest caffe master branch] from (https://github.com/BVLC/caffe), it seems to merge successfully without any conflicts. and then i make -j8 && make pycaffe -j8, it still looks good. But, when i run the demo in tools/demo.py of faster-rcnn, it got some problem like this:

I0701 16:42:13.574218 2278 net.cpp:156] Memory required for data: 117496524
I0701 16:42:13.574223 2278 layer_factory.hpp:77] Creating layer proposal
I0701 16:42:13.583842 2278 net.cpp:91] Creating Layer proposal
I0701 16:42:13.583858 2278 net.cpp:425] proposal <- rpn_cls_prob_reshape
I0701 16:42:13.583864 2278 net.cpp:425] proposal <- rpn_bbox_pred
I0701 16:42:13.583869 2278 net.cpp:425] proposal <- im_info
I0701 16:42:13.583875 2278 net.cpp:399] proposal -> rois
Traceback (most recent call last):
File "demo.py", line 135, in
net = caffe.Net(prototxt, caffemodel, caffe.TEST)
AttributeError: can't set attribute

did you ever seen this before?
i don't find out what the problem it is!

@rbgirshick can you do me a favor?
thanks!

[Austriker]

@zimenglan-sysu-512 I have a working py-faster-rcnn with an up-to-date caffe :
Austriker/py-faster-rcnn

[stonycat]

@zimenglan-sysu-512 I meet the same problem, have you solved it ?

[zimenglan-sysu-512]

@stonycat hi, after merge, i just Remove "self_.attr("phase") = static_cast<int>(this->phase_);" from include/caffe/layers/python_layer.hpp after merging.

[stonycat]

@zimenglan-sysu-512 thank you for your reply.Have you met conflicts when you merge caffe master branch into caffe-fast-rcnn?

[zimenglan-sysu-512]

hi @stonycat i don't have any conflicts.

[mdadhich]

@Austriker if I just clone faster_rcnn from your git, do I have to build everything I have done from Ross's. and will it work if I build it using cmake and not make as I am using Red-Hat. Thanks
Please have a look, my problem is described here - rbgirshick/py-faster-rcnn#494

[soulslicer]

Made it work with pascal:

https://github.com/soulslicer/caffe-fast-rcnn

Just change proposal_layer.py

     def setup(self, bottom, top):
         # parse the layer parameter string, which must be valid YAML
-        layer_params = yaml.load(self.param_str_)
+        layer_params = yaml.load(self.param_str)
 
         self._feat_stride = layer_params['feat_stride']
         anchor_scales = layer_params.get('scales', (8, 16, 32))
@@ -61,7 +61,7 @@ class ProposalLayer(caffe.Layer):
         assert bottom[0].data.shape[0] == 1, \
             'Only single item batches are supported'
 
-        cfg_key = str(self.phase) # either 'TRAIN' or 'TEST'
+        cfg_key = 'TEST' if self.phase == 1 else 'TRAIN'