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

Recent progresses on self-supervised 3D human action representation learningare largely attributed to contrastive learning. However, in conventionalcontrastive frameworks, the rich complementarity between different skeletonmodalities remains under-explored. Moreover, optimized with distinguishingself-augmented samples, models struggle with numerous similar positiveinstances in the case of limited action categories. In this work, we tackle theaforementioned problems by introducing a general Inter- and Intra-modal MutualDistillation (I$^2$MD) framework. In I$^2$MD, we first re-formulate thecross-modal interaction as a Cross-modal Mutual Distillation (CMD) process.Different from existing distillation solutions that transfer the knowledge of apre-trained and fixed teacher to the student, in CMD, the knowledge iscontinuously updated and bidirectionally distilled between modalities duringpre-training. To alleviate the interference of similar samples and exploittheir underlying contexts, we further design the Intra-modal MutualDistillation (IMD) strategy, In IMD, the Dynamic Neighbors Aggregation (DNA)mechanism is first introduced, where an additional cluster-level discriminationbranch is instantiated in each modality. It adaptively aggregateshighly-correlated neighboring features, forming local cluster-levelcontrasting. Mutual distillation is then performed between the two branches forcross-level knowledge exchange. Extensive experiments on three datasets showthat our approach sets a series of new records.

I$^2$MD: 3D Action Representation Learning with Inter- and Intra-modal Mutual Distillation