Questions about the deform_psroi_demo.py

[carrierlxk]

In deform_psroi_demo.py, I have two questions about the details: 1.Why the offset of bounding box(red box) comes from output of rfcn_cls_offset layer rather than the output of rfcn_bbox_offset layer? I think the latter is more related to sub bbox location, or because the rfcn_cls_offset is related to foreground object? 2.Why set the value of trans_std to 0.1 in function show_dpsroi_offset? Thank you!!

[chengshuai]

Hi @carrierlxk

I have compile the mxnet and copy the folder mxnet/pyhton to external/mxnet,,then run the command python ./rfcn/demo.py ,the error is File "./rfcn/demo.py", line 28, in from core.tester import im_detect, Predictor File "/home/chengshuai/test_sample/Deformable-ConvNets-master/rfcn/core/tester.py", line 15, in from module import MutableModule File "/home/chengshuai/test_sample/Deformable-ConvNets-master/rfcn/core/module.py", line 19, in from mxnet.initializer import Uniform, InitDesc ImportError: cannot import name InitDesc I do not know the why the error hanppen and could you tell me the deformal convoluthon network process.

Thank you!

[gd-zhang]

@carrierlxk

rfcn_cls_offset is learnt to enhance classification branch while rfcn_bbox_offset tends to strengthen localization. In this sense, rfcn_cls_offset is likely to focus on objects and rfcn_bbox_offset will partly move out of the box to see more context. That's why we visualize rfcn_cls_offset.

Hope this helps.