help !!! about apap-image-stitching

[EadCat]

Hello.
Do you mean the benchmarking in section 4.2 in the paper?

[czy2134]

yes

[EadCat]

After the RANSAC, you can use the SIFT matching point between the source and destination image.
Of course, you can split these matching points randomly by modifying the source code.

But I'm not sure what "train" and "test" mean in the paper.

I guess after warping the matching points of the source image using APAP, they measured the distances of the matching points (of test set) between src & dst images within the stitched output.

Note that I used OpenCV for the SIFT instead of VLFeat, and this program stitches inputs recursively (to alleviate perspective distortions).

[czy2134]

Thanks for your reply.
I don't know after warping how to get the matching points of the source image using APAP.

Can I re-extract feature points from the warped image by SIFT?

[EadCat]

I didn't consider quantitative evaluation, so it is impossible to extract the warped points with this source code.
You need to modify the code to track the warped matching points. The class responsible for APAP warping may be modified to store feature point coordinates.