how to generate proposal?

[hengck23]

I was running the demo code:
https://github.com/daijifeng001/R-FCN/blob/master/experiments/script_rfcn_demo.m

the proposal are loaded from file:
proposals = load(fullfile(demo_dir, [im_names{j}, '_boxes.mat']));

How can I run the detection of one image end-to-end?

Thank you very much

[daijifeng001]

Hi, then you need to follow the 4-step training in paper to train a region proposal network and a R-FCN network, and to share features between them.

[hengck23]

@daijifeng001
Thank you for the reply. Can I confirm that the demo code do not provide end-to-end detection at test time? Becuase in faster-rcnn demo, we can just input an image and both proposal and final detection can be output.

(note: I am referring to detection and not training. I want to test my own images but I do not have their proposal mat files)

[daijifeng001]

The demo does not provide proposal generation, but it is very fast in inference and would not be a trouble for computation.

[daijifeng001]

Can you try the dropbox link?

From: hengck23 [mailto:notifications@github.com]
Sent: Thursday, July 21, 2016 3:33 PM
To: daijifeng001/R-FCN R-FCN@noreply.github.com
Cc: Jifeng Dai daijifeng001@gmail.com; Mention mention@noreply.github.com
Subject: Re: [daijifeng001/R-FCN] how to generate proposal? (#10)

@daijifeng001 https://github.com/daijifeng001
Thank you for the reply. Can I confirm that the demo code do not provide end-to-end detection at test time? Becuase in faster-rcnn demo, we can just input an image and both proposal and final detection can be output.

(note: I am referring to detection and not training. I want to test my own images but I do not have their proposal mat files)

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub #10 (comment) , or mute the thread https://github.com/notifications/unsubscribe-auth/AH9IlcbROlYFyKggzS8OyJry3hYKc2L0ks5qXyDIgaJpZM4JRhFM . https://github.com/notifications/beacon/AH9IlRTMNVd9bCw0U6UhEE5VRHRwsv0tks5qXyDIgaJpZM4JRhFM.gif

[hengck23]

@daijifeng001
I can download the model at "Demo R-FCN model: DropBox, BaiduYun"

If these models already share features for RPN and frcnn, then I can write the code for end-to-end detection and share it here. Currently I am convert your script_rfcn_demo.m to C++ and run the demo in windows C++ caffe only. I convert the proposal mat files to text file so that it can be read by c++.
This is now OK.

Let try to write some code using demo model to make proposals and get back to you later.
Do I use the same anchor boxes as faster RCNN? (i.e. 9 aspect ratio at each location)
Thanks!

[daijifeng001]

Yes.

From: hengck23 [mailto:notifications@github.com]
Sent: Thursday, July 21, 2016 4:24 PM
To: daijifeng001/R-FCN R-FCN@noreply.github.com
Cc: Jifeng Dai daijifeng001@gmail.com; Mention mention@noreply.github.com
Subject: Re: [daijifeng001/R-FCN] how to generate proposal? (#10)

@daijifeng001 https://github.com/daijifeng001
I can download the model at "Demo R-FCN model: DropBox, BaiduYun"

If these models already share features for RPN and frcnn, then I can write the code for end-to-end detection and share it here. Currently I am convert your script_rfcn_demo.m to C++ and run the demo in windows C++ caffe only. I convert the proposal mat files to text file so that it can be read by c++.
This is now OK.

Let try to write some code using demo model to make proposals and get back to you later.
Do I use the same anchor boxes as faster RCNN? (i.e. 9 aspect ratio at each location)
Thanks!

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub #10 (comment) , or mute the thread https://github.com/notifications/unsubscribe-auth/AH9IlZU_oNCrLSMcEYuhAOSpv4PvOJtQks5qXyyqgaJpZM4JRhFM . https://github.com/notifications/beacon/AH9IlXZvFlZ7zUHcGdza4XywITCGDq02ks5qXyyqgaJpZM4JRhFM.gif

[zzningxp]

Should I using the same way like in faster-rcnn to get proposals in a different function using a different caffe prototxt define network and model? Or, just using the current model, but how?
@hengck23 @daijifeng001