search

PruneTruong / DenseMatching

Dense matching library based on PyTorch
GNU Lesser General Public License v2.1
690 stars 82 forks source link

Used kernel for WarpCRANSACflow #23

Closed Camilochiang closed 1 year ago

Camilochiang commented 2 years ago

Hei! Great code and organized github! Thanks for your work ;) I'm using your weights to align images using RANSAC-flow. The only problem that I have is that i'm getting some deformation in the images. I look around your code and in your WarpC paper bud didn't find any detail about the size of the used kernel. In ransac-flow was 7. Did you use the same size?

Thanks!

PruneTruong commented 2 years ago

Hi, thanks for your interest in our work! I think I used kernel 7 for RANSAC-Flow, along with all default parameters. What kind of deformations are you seeing? Did you try reproducing the results of the paper with the provided weights?

Camilochiang commented 2 years ago

Yes. Got same or really similar results, but when testing in my images, there are deformations For example: This is the target image (To which one I want to align) IMG_000X__1

Here the raw image to be aligned (to the previous image. The white part is because it is an screenshot) Screenshot from 2022-10-11 08-54-54

And after the alignment process IMG_000X__3

there are some image deformations ...

Thanks!

PruneTruong commented 2 years ago

Hi, this indeed sounds pretty weird, almost like it is shifted in the wrong direction. I do not have access to RANSAC-Flow codebase anymore, but did you check how 1) RANSAC-Flow original is doing, 2) the first stage homography is doing. If the homography is failing, then there is no way the network can recover since it assumes small displacement. Otherwise, is there any particular reason you want to use RANSAC-Flow? PDC-Net should be doing ok on these kind of examples i think.

PruneTruong commented 1 year ago

I am closing due to lack of recent activity. Feel free to reopen if you still have a question.