AI Paper Atlas

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

DySeT: a Dynamic Masked Self-distillation Approach for Robust Trajectory Prediction

Mozghan Pourkeshavarz, Arielle Zhang, Amir Rasouli

ECCV 2024 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Self-supervision in supervised pipelines · 8 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
replace random masking with dynamic informativeness-based token sampling (ablation)masked trajectory autoencoder · Argoverse2 valminFDE6: 1.486 -> 1.306Improved
add masked self-distillation from EMA teacher (ablation)DySeT · Argoverse2 valminADE6: 0.751 -> 0.724Improved
set distillation weight lambda too high (10) (ablation)DySeT · Argoverse2 valminADE6: 0.724 -> 0.804Hurt
use dynamic sampling at high masking ratio (70%) versus random sampling (ablation)DySeT · Argoverse2 valminADE6: minADE 0.724, minFDE 1.278Improved
DySeT versus Forecast-MAE in unseen cities (finetune)DySeT · Argoverse2 val unseen citiesminFDE6: 1.613 -> 1.328Improved
DySeT versus Forecast-MAE under targeted adversarial perturbationDySeT · Argoverse2 valPRS (%): 78.51 -> 86.83Improved
DySeT versus GP-Graph and GroupNet on UNIV source domainDySeT · ETH/UCY unseen domains, UNIVminADE20: 0.61 vs GP-Graph 0.54Hurt
DySeT versus Forecast-MAE finetune on single-modal metricDySeT · Argoverse2 testminFDE1: 4.35 -> 4.41Hurt

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