AI Paper Atlas

Research map / Video / Action recognition

MoLo: Motion-Augmented Long-Short Contrastive Learning for Few-Shot Action Recognition

Xiang Wang, Shiwei Zhang, Zhiwu Qing, Changxin Gao, Yingya Zhang, Deli Zhao, Nong Sang

CVPR 2023 · Video › Action recognition · approach: Disentangled action recognition · 101 citations (OpenAlex)

This paper tried a common change: Add a contrastive loss

add the long-short contrastive objective to Bi-MHM · SSv2-Full, 5-way 1-shot · accuracy: 44.6 -> 52.2 Improved

What 1,317 accepted papers report for this change

See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the long-short contrastive objective to Bi-MHM (ablation)Bi-MHM · SSv2-Full, 5-way 1-shotaccuracy: 44.6 -> 52.2Improved
add the motion autodecoder to Bi-MHM (ablation)Bi-MHM · SSv2-Full, 5-way 1-shotaccuracy: 44.6 -> 53.2Improved
combine long-short contrastive learning and the motion autodecoder (ablation)Bi-MHM · SSv2-Full, 5-way 1-shotaccuracy: 44.6 -> 56.6Improved
replace frame-difference reconstruction target with RAFT flow (ablation)MoLo · SSv2-Full, 5-way 1-shotaccuracy: 56.6 -> 56.8Improved
add the motion autodecoder (ablation)TRX · SSv2-Full, 5-way 1-shotaccuracy: 42.0 -> 45.6Improved
increase the temporal Transformer depth from one to two layers (ablation)MoLo · Kinetics, 5-way 1-shotaccuracy: 74.0 -> 72.5Hurt
increase the temporal Transformer depth from one to three layers (ablation)MoLo · Kinetics, 5-way 1-shotaccuracy: 74.0 -> 71.6Hurt
replace the learnable global token with temporal average pooling (ablation)MoLo · SSv2-Full, 5-way 1-shotaccuracy: 56.6 -> 54.8Hurt
remove the long-short contrastive loss and use Transformer-only matching (ablation)MoLo · SSv2-Full, 5-way 1-shotaccuracy: 56.6 -> 53.2Hurt
use cross-video rather than within-video contrastive learning (ablation)MoLo · SSv2-Full, 5-way 1-shotaccuracy: 54.8 -> 56.6Improved
use cross-video rather than within-video contrastive learning (ablation)MoLo · Kinetics, 5-way 1-shotaccuracy: 73.9 -> 74.0No clear effect
use deeper ResNet backbones (ablation)MoLo · SSv2-FullaccuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy