End-to-end CNN detection and description
Published in
Aug 5, 2021
Let’s explore using convolutional networks to solve problems stemming from detecting image points.
The LIFT network includes the sequential stages of first detection, then orientation estimation, and finally descriptor computation.
The network is trained in parts, starting with positive examples for training the descriptor, then the descriptor for further training the orientation evaluator, and at the end, the descriptor and orientation evaluator are both utilized for training the detector.