Project 2 - Panorama
Minh Quan Nguyen and Khanh Tran
CSE 455: Computer Vision
Artifact
Gasworks Park: Mounted
We were unable to get signification portions of our project working completely correctly so both our panorama images were generated using the solution executables. This is the view from the top of the hill at Gasworks Parks. Unfortunately, the blending of the city in the distance didn't turn out that well. We had a version taken at with a more zoomed in view, but the Kaiden head gave us trouble and the stitching together of that sequence proved unsuccessful.
My Backyard: Handheld
This a the handheld sequence taken of my backyard. There are some blending issues with the trees in the middle. We think it is because there wasn't enough overlap between the two images so not enough features were matched.
Test sequence
Here's what our code generated for the panorama. It's quite aweful and we regret not spending more time on the project in order to get everything working.
ROC comparison of our code vs. SIFT
Obviously, our feature descriptor code was buggy which produced some very strange graphs. However, it looks like our feature detection worked. The plots, seem to indicate that we have many flase positives, but with incorrect feature descriptor values, they're more likely meaningless.
Same as above. The graphs aren't correct, but it looks like the feature detection is.