Published on:
8 November 2023
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Hezhen Hu,
Xiaoyi Dong,
Jianmin Bao,
Dongdong Chen,
Lu Yuan,
Dong Chen,
Houqiang Li
PersonMAE pre-trains a model for person re-ID using masked autoencoders
The model learns to predict masked regions from visible ones, handling occlusion
This teaches multi-level awareness of visual details and semantics
PersonMAE gets SOTA results on supervised and unsupervised re-ID tasks
Person re-identification with masked autoencoders
This paper proposes a pre-training framework called PersonMAE that uses masked autoencoders to learn useful representations for person re-identification. The model is trained to predict masked regions of an image from visible regions, which teaches it to handle occlusion and focus on multiple levels of detail. In experiments, PersonMAE achieves state-of-the-art performance on supervised and unsupervised re-ID benchmarks.
Masked modeling for understanding motion in crowded scenes
Self-supervised video representation learning using masked reconstruction
Diffusion Models as Masked Autoencoders: A New Approach for Learning Visual Representations
Self-Supervised Learning for Functional Connectivity Networks
Enhancing Monocular 3D Detection via Feature Masking and Completion
Reconstructing complete 3D shapes from partial views
No comments yet, be the first to start the conversation...
Sign up to comment on this paper