low match ratio when images have large rotation change

[minxuanjun]

No description provided.

[minxuanjun]

@atakagi-fixstars
Thank your great work! When I test your code in an image pair with a large rotation change, it has a low match ratio w.r.t original efficient feature descriptor. Can you help me?

[atakagi-fixstars]

Hi, @minxuanjun

Thank you for your information.
I'll check it.

I would appreciate it if you could tell me more about the situation.
For example, input images or rotation degrees, etc.

[minxuanjun]

Hi, @minxuanjun

Thank you for your information. I'll check it.

I would appreciate it if you could tell me more about the situation. For example, input images or rotation degrees, etc.
@atakagi-fixstars Thank you for your reply. you can test below the image pair.

[atakagi-fixstars]

Hi, @minxuanjun

I found a mistake in keypoint angle calculation.
In the angle calculation, conversion to degree was required, but it was not done.
This problem seems to be solved by moving convertToDegree into calcAnglesKernel.

I will fix it soon.

[atakagi-fixstars]

Hi, @minxuanjun

Thank you for your great contribution. We fixed it.
Changelog: 1.0.0...1.0.1

We have confirmed that the match ratio has recovered.
For example, when I ran the sample with the command below, I found that the number of matches had recovered from approximately 25% to 55%.

./samples/sample_feature_matching image1.png image2.png --nonmax-radius=5 --descriptor-type=1

before

=== configulations ===
image size      : [669 x 448] and [664 x 440]
descriptor type : HashSIFT
descriptor bits : 256
max keypoints   : 10000

=== extract features ===
number of keypoins: 3035 2815
=== match features ===
number of matches: 741

after

=== configulations ===
image size      : [669 x 448] and [664 x 440]
descriptor type : HashSIFT
descriptor bits : 256
max keypoints   : 10000

=== extract features ===
number of keypoins: 3035 2815
=== match features ===
number of matches: 1615