pengxj / action-faster-rcnn

59 stars 22 forks source link

how the multi-region is implemented in the code? #6

Open helxsz opened 7 years ago

helxsz commented 7 years ago

Hi,@pengxj, thanks for the work, I just reviewed the paper on the multi-region part, which has 4 parts including the whole body, upper body, bottom body, but I don't know how is the 4th region defined, excluding the body part? Additionally for the multi-region implementation, I thought it would be implemented directly on the roi_data_layer/minibatch.py or roi_data_layer/layer.py, but I didn't find there is connection, could you tell me where this part is being implemented ?

pengxj commented 7 years ago

Hey, That's wrong. The multi-region part is based on the RPN proposals. All the others are the same as normal except for the proposals from RPN which are replaced by part regions, see Fig.3.

helxsz commented 7 years ago

Thanks, pengxj. Could you point out in the code where it is being implemented, pls ?

pengxj commented 7 years ago

There is no implementation in my github. I don't recommend the multi-region part in my paper since it only improves a little bit while a more powerful network (i.e., resnet) can easily outperform them.

On Mon, Jun 26, 2017 at 4:48 PM, xsz notifications@github.com wrote:

Thanks, pengxj. Could you point out in the code where it is being implemented, pls ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/pengxj/action-faster-rcnn/issues/6#issuecomment-311081784, or mute the thread https://github.com/notifications/unsubscribe-auth/AHPPqJ5jQFEtfzPeoI46InKT9hb73n_6ks5sH8S7gaJpZM4OEUgY .

-- Best Regards!

Xiaojiang Peng Position: PostDoc CVL, Idiap Switzerland Cell: +41 7 7520 7524 https://www.idiap.ch/~xpeng

wangwen39 commented 6 years ago

hey thanks for sharing the code, could you please share the code of multi-region ? and i just use this in my research. thanks so much!