Test Sequence

The following panorama represents the result of stitching the building images together, i.e. warp08 through warp11.tga: The full-size image is linked here. Additionally, here is the viewer version of it:

Handheld Sequence

The following panorama represents the result of stitching of a bunch of images taken inside my apartment living room: The full-size image is linked here. Additionally, here is the viewer version of it:

Handheld sequence with homography

Strengths and weaknesses

We are fairly confident in the correctness of our warping and align functions. Output comparisons to that of the solution have yielded sufficiently conclusive evidence that this is case. Blending works as expected. One issue we encountered early on was the appearance of artifacts in the form of these cyclically occurring regions of transparent/dark rectangles along the top and bottom. The issue was that the function was blending black pixels, which comprised the boundary of the spherically unwarped image, with the pixels from the actual image. We remedied this by looking at all the pixels used by the Lerp function and filtering out black ones.

The other issue was that of cropping. We did the cropping by removing the leftmost image and then translating accordingly, whereas we should have cropped it, cut the rightmost image down the middle, and put the right half on the left side of the image.

The other issue is that the cropping isn't occuring correctly. It correctly removes the leftmost image, which is duplicated on the right side.

Another observation is that with the building panorama, there are noticeable differences in exposure.