Can I compute the rectangle image not the square image?

[G-YY]

I try to test the 1280*1024 image ,but the result is not correct,Thank you

[orzzzjq]

Cannot compute a rectangle image directly due to efficient consideration. But you can simply put the rectangle image into a large square image. Or maybe you can try to modify the transpose operation in the code, to let it work for rectangle images, but it's a little difficult.

[y123l]

The new version of the code without transpose operation seems to handle the rectangle image wrong as well, do you have any suggestions

[orzzzjq]

You are right, I didn't implement a version that handles rectangle images. The new version performs an implicit transpose of the image. However, the code can be modified to support rectangle inputs.

The current method divides the image into small squares and does some computation within the small squares, then writes the result to the transposed position of the square. If we divide a rectangle image into small pieces of squares and do the implicit transpose for the rectangle image correspondingly, then we can support rectangle images as well.

Unfortunately, currently I don't have time to implement it. Someone achieved this using the above method, even for 3D PBA, but they may not be willing to share due to commercial considerations.

You may try to do so if you really want to handle rectangle images. It would be much easier for 2D PBA. Otherwise, for now, you may simply put the image into a large square and pass it to PBA. It won't be too slow if the rectangle image is well-shaped (close to a square).

[y123l]

Thank you for your detailed explanation. I will consider trying to adapt the method for rectangular inputs myself.