AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Uncovering the Hidden Dynamics of Video Self-supervised Learning under Distribution Shifts

Pritam Sarkar, Ahmad Beirami, Ali Etemad

NeurIPS 2023 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Pretext-task visual pre-training · 2 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
finetune instead of linear evaluationVSSL video encoders · InD averageaccuracy: 63.2% -> 71.9%Improved
finetune instead of linear evaluationVSSL video encoders · OoD averageaccuracy: 31.5% -> 37.4%Improved
contrastive pretraining (negatives) vs supervised on multiple concurrent shiftsV-SimCLR vs V-Supervised · Sims4Action top-down+synthetic, finetunedaccuracy: 44.3 -> 63.1Improved
use masked autoencoding instead of contrastiveV-MAE vs V-SimCLR · CharadesEgo ego viewpoint, linearmAP: 12.7 -> 10.9Hurt
use V-MAE on low-resolution surveillanceV-MAE vs V-Supervised · TinyVirat-v2 FT OoDtop-1 accuracy: 24.5 -> 29.1Improved
finetune HMDB-trained model vs frozen encoder (reversed source shift)V-BYOL · HMDB to UCF source shiftaccuracyHurt
end-to-end finetune contrastive encoders under temporal perturbationscontrastive/non-contrastive VSSL · temporally perturbed inputsrobustnessHurt
use frozen contrastive encoders for open-setV-SimCLR vs V-DINO · Open-set UCF101/HMDB51 linearAUC: 79.4 -> 49.9Hurt
finetune contrastive models for open-set recognitionV-MAE vs V-SimCLR · Open-set K400/UCF finetunedAUC: 55.1 -> 63.0Improved
aggressive cropping [0.08,1.0] for contrastive methods (ablation)contrastive VSSL · UCF101 linear SVMaccuracyImproved
aggressive cropping for masked autoencoding (ablation)V-MAE · UCF101accuracyHurt
remove temporal information (single frames) (ablation)V-MAE · UCF101 linearaccuracy: 76.2 -> 5.6Hurt

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