Is it possible to train a neural network or a classifier on SIFT keypoints and descriptors? I am working on a project which involves detecting an object (a bottle of energy drink) in an image and then detect keypoints and descriptors of that object. Then take another image and repeat the process, and once the descriptors and keypoints are detected, they are matched with the descriptors detected in previous step using FLANN matcher. What I am seeking now is that: is it possible to train a neural network on these descriptors with a class assigned to them and instead of FLANN matcher use that trained model. So that In second step, instead of using FLANN, I could just pass the descriptors to the trained model and get the class?
Asked
Active
Viewed 93 times
1 Answers
1
Yes, it is possible. You just need to make sure that the feature vectors/descriptors have the same size, as neural networks and other ML methods require you to fixed size features, they cannot be variable sized.
Also I think these features are not so good, CNN features probably work much better.
Dr. Snoopy
- 1,142
- 6
- 12