English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Model pre-training is essential in human-centric perception. In this paper,we first introduce masked image modeling (MIM) as a pre-training approach forthis task. Upon revisiting the MIM training strategy, we reveal that humanstructure priors offer significant potential. Motivated by this insight, wefurther incorporate an intuitive human structure prior - human parts - intopre-training. Specifically, we employ this prior to guide the mask samplingprocess. Image patches, corresponding to human part regions, have high priorityto be masked out. This encourages the model to concentrate more on bodystructure information during pre-training, yielding substantial benefits acrossa range of human-centric perception tasks. To further capture humancharacteristics, we propose a structure-invariant alignment loss that enforcesdifferent masked views, guided by the human part prior, to be closely alignedfor the same image. We term the entire method as HAP. HAP simply uses a plainViT as the encoder yet establishes new state-of-the-art performance on 11human-centric benchmarks, and on-par result on one dataset. For example, HAPachieves 78.1% mAP on MSMT17 for person re-identification, 86.54% mA on PA-100Kfor pedestrian attribute recognition, 78.2% AP on MS COCO for 2D poseestimation, and 56.0 PA-MPJPE on 3DPW for 3D pose and shape estimation.

HAP: Structure-Aware Masked Image Modeling for Human-Centric Perception