AI Paper Atlas

Research map / Video / Action recognition

ASCNet: Self-Supervised Video Representation Learning With Appearance-Speed Consistency

Deng Huang, Wenhao Wu, Weiwen Hu, Xu Liu, Dongliang He, Zhihua Wu, Xiangmiao Wu, Mingkui Tan, Errui Ding

ICCV 2021 · Video › Action recognition · approach: Contrastive video representation learning · 47 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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 self-supervised ACP pretraining versus training from scratch (ablation)3D R18 · UCF-101 (Mini-K200)top-1 accuracy: 42.40% -> 64.76%Improved
add SCP task on top of ACP (ablation)ASCNet 3D R18 · UCF-101 (Mini-K200)top-1 accuracy: 64.76% -> 70.71%Improved
use SCP alone without ACP (ablation)3D R18 · UCF-101 (Mini-K200)top-1 accuracy: 42.40% -> 43.40%No clear effect
use speed consistency instead of speed prediction (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 68.93% -> 70.71%Improved
use appearance-based retrieval for SCP pairs instead of random different instances (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 69.55% -> 70.71%Improved
sample SCP clips from similar instances instead of same instance (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 69.20% -> 70.71%Improved
use different playback speeds instead of both at 1x (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 64.52% -> 70.71%Improved
use speed set 4x,8x instead of 1x,2x (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 70.71% -> 72.16%Improved
increase pretraining batch size from 256 to 1024 (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 72.13% -> 72.20%No clear effect
stack Gaussian blur, grayscale and solarization on color jittering (ablation)ASCNet · UCF-101 (Mini-K200)top-1 accuracy: 62.43% -> 72.16%Improved
pretrain longer, 100 to 400 epochs (ablation)ASCNet 3D R18 · UCF-101 (K400)top-1 accuracy: 76.34 -> 81.50Improved
apply ASC pretraining versus random init (ablation)R(2+1)D · UCF-101 (Mini-K200)top-1 accuracy: 56.00% -> 75.95%Improved

Later papers that compared with this method

1 later accepted paper reports ASCNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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