Well done: Overall the blending is awesome

Not well: There aren't really any glaring problems, though it is unfortunate that the edges of the photos are cut off since it makes the panorama rather short overall.

no extra credit done

The ROC comparison graph shows a ratio of false positive rates versus versus true positive rates as the threshold is varied for the given feature matching methods. In general, we want more area under the graph such that the graph is towards the left where there are more true positive matches. In light of this, our plot shows better performance by SIFT over our MOPS feature detection and matching, though MOPS does fairly well. We also see a significant improvement after applying the ratio test versus SSD for scoring the matches, as we would hope to see.

Again, we see the same sort of performance comparison between our feature matching and SIFT, though with rather poorer performance overall.

To detect features in the images we used the Harris operator on the image to create maps like these, choosing local maxima as features. This is the resulting Harris scores over the pixels in the image Yosemite1.jpg (above), and for img1.ppm from the "graf" image set (below), next to their original images.

Worked well: Blending is overall well done.

Not well: The last images are not blended well. We guess that it couldn't find distinct features because they only have straight vertical lines. We also weren't sure which camera we had used so it seems we used the wrong focal length and radial distortion coefficients when warping, causing both the images in this sequence and the next not to be correctly distorted into spherical coordinates for blending.

We didn't do any extra credit

Worked well: Like the first panorama, blending is overall well

Not well: Some images are not blended well. This one also might not have distinct features to be blended smoothly.

We didn't do any extra credit