It loosk to me that its ORB extractor has a different implementation than that from OpenCV. Is it a different variant of ORB extractor?
When I tested with monokitti, at frame 204, using ORB & knnmatch opencv code, it seems to fail, has very few descriptor e.g. <5, whereas ORB SLAM2 code here, still have matches like 60s. Why is that? Is it ORB descriptor code here performs better than opencv orb extractor?
It loosk to me that its ORB extractor has a different implementation than that from OpenCV. Is it a different variant of ORB extractor?
When I tested with monokitti, at frame 204, using ORB & knnmatch opencv code, it seems to fail, has very few descriptor e.g. <5, whereas ORB SLAM2 code here, still have matches like 60s. Why is that? Is it ORB descriptor code here performs better than opencv orb extractor?