AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights.
Byeongho HeoSanghyuk ChunSeong Joon OhDongyoon HanSangdoo YunGyuwan KimYoungjung UhJung-Woo HaPublished in: ICLR (2021)
Keyphrases
- scale invariant
- keypoints
- scale space
- affine invariant
- natural images
- scale invariance
- multiscale
- linear combination
- feature transformation
- object class recognition
- textured surfaces
- local binary pattern
- interest point detectors
- image structure
- texture images
- learning rate
- face recognition
- figure ground separation
- invariant features
- translation and scale invariant
- gaussian derivatives
- image matching
- denoising
- object detection
- object recognition
- neural network