I found this code for evaluating the SIFT descriptor w.r.t. viewpoint changes using the graffiti data set (present in OpenCV too).
I have two questions:
- Supposing that I want to test the performance of another dataset, I would need an homography as ground truth. How can I compute it?
- The code above is specifically designed for the SIFT vector (which distance is measured through euclidean distance, if I'm not wrong). What if I want to evaluate a binary descriptor (where the distance is measured through hamming distance)?
- There is any built-in tool in OpenCV for descriptors evaluation?