Open KinglittleQ opened 5 years ago
When training with float point descriptors, the K-means step may be not converging. Adding a limitation of max iteration times could solve it.
Can this solve floating-point training problems such as sift, surf, etc? Is that what many people care about when the matching score is zero?
When training with float point descriptors, the K-means step may be not converging. Adding a limitation of max iteration times could solve it.