Open tompollok opened 4 years ago
That would be great. How would the process work, should I create a pull request?
I think all you have to do is to create a pullrequest containing your aff_*.cpp and hpp files and maybe restucture it given the opencv guidelines.
I just checked: @Yosshi999 is working on this as a GSoC (Google Summer of Code).
https://github.com/Yosshi999/opencv/commits/gsoc_sift-add-asift
I dont know how he will implement it compared to yours. But maybe your solution is still an added valuable because you can affinely match any features. So I guess one way would be to get in touch directly with Yosshi999 or just making a initial PR without too much refactoring and then seeing the discussion feedback.
Also OpenCV 4.4 is to be released soon. Otherwise the next release always takes 2-3 months.
@tompollok I think it's better if I contact @Yosshi999. Otherwise, I'm afraid there will be competing pull requests.
I found the person's email in the log message in his/her commits. I'll try to reach him/her there.
Hello, @kukuruza. It's OK to discuss here or in my draft PR https://github.com/terfendail/opencv/pull/9 I referenced your code in implementing asift in C++, but actually it became far from yours because I thought we need tilt/roll stepsize as parameters, and some missing features (masking, Gaussian blur after tilting, etc.)
Hey @Yosshi999, makes sense. Is there any functionality in this repo that is not in your draft PR? It seems that you are implementing everything that's needed. The only thing not mentioned is progressive matching (tilt the image as little as necessary)
That's right, I haven't implemented affine matcher yet.
I think the order of class_id
in my detector is compatible with yours, but my detector may exceed MaxPossibleNumViews because tilt/roll stepsize are parameterized.
Okay, then I think your repo will duplicate all the necessary functionality?
In terms of transplantation of affine_detect
in https://github.com/opencv/opencv/blob/3.4/samples/python/asift.py , I think I've completed necessary features.
Is it okay for you if this work is pushed into opencv?