Have some question for replying~

[lancexiang]

1.R[k]["size"]= len(masked_pixels / 4)
Why there division by 4? It's seems no influence for the len function.
2.skimage.feature.local_binary_pattern(img[:, :, colour_channel], 8, 1.0)
hist = numpy.concatenate([hist] + [numpy.histogram(fd, BINS, (0.0, 1.0))[0]])
The output range of LBP feature is (0.0, 255.0), then why the histogram range is (0.0, 1.0), that means
leave off the pixel > 1? or maybe a bug?

[Page-David]

Hi, the 4 seems to be a magic number. And as for the second problem, it seems to be a duplicate of issue #21. I think you are correct, but as the authors have no activity for months in this repo, I do not have expectations that these bug will be fixed.

[lancexiang]

Thank you for your reply. I ever think there maybe a wrong typewrite about len(masked_pixel / 4), or the right is len(masked) / 4 for calculate the side's length of the square. But that's not reasonable, for the rectangle of region proposal is not always square~

[Mickeybaby]

Because the author has abandoned the repo.So I come to ask for your help.
I want to ask does the order of the final returned results follow the ascending order of the scores?
How can I see the score information in this program?
Best wishes!

[Mickeybaby]

@Page-David @lancexiang
I want to ask does the order of the final returned results follow the ascending order of the scores?
How can I see the score information in this program?
And in fast rcnn,We usually use 2,000 proposals, and how do we select these 2,000 proposals? Take this program as an example.
I am a beginner in object detection.
Thank you!Best wishes!

[Page-David]

@Mickeybaby The algorithm is based on similarity between regions, so I am not sure what you mean by “scores”. For the question about fast rcnn, I still know little about it...