AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Encoding Crowd Interaction With Deep Neural Network for Pedestrian Trajectory Prediction

Yanyu Xu, Zhixin Piao, Shenghua Gao

CVPR 2018 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 293 citations (OpenAlex)

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
Use CIDNN to model crowd interactions and predict displacementpedestrian trajectory prediction · ETH, HOTEL, ZARA 1, ZARA 2, UCY, GC, CUHK Crowd, subway stationADE: ETH: 0.35 -> 0.09; HOTEL: 0.11 -> 0.11; ZARA 1: 0.20 -> 0.15; ZARA 2: 0.23 -> 0.10; UCY: 0.25 -> 0.12; GC: 0.020 -> 0.01Improved
Predict displacement rather than coordinates directly (ablation)CIDNN · ETH, ZARA 2, GCADE: ETH: 0.23 -> 0.09; ZARA 2: 0.24 -> 0.10; GC: 0.018 -> 0.012Improved
Add the crowd interaction module (ablation)CIDNN · ETH, ZARA 2, GCADE: ETH: 0.11 -> 0.09; ZARA 2: 0.15 -> 0.10; GC: 0.055 -> 0.012Improved
Use learned location affinity instead of Gaussian distance affinity (ablation)CIDNN · ETH, ZARA, GCADE: ETH: 0.11 -> 0.09; ZARA: 0.16 -> 0.10; GC: 0.055 -> 0.012Improved
Use an LSTM motion encoder instead of neighboring-frame displacements (ablation)CIDNN · ETH, ZARA, GCADE: ETH: 1.37 -> 0.09; ZARA: 0.17 -> 0.10; GC: 0.018 -> 0.012Improved
Use a two-layer rather than one-layer or three-layer stacked LSTM (ablation)CIDNN motion encoder · GCMSE: 0.014, 0.0125, and 0.013Improved
Use coordinates rather than neighboring-frame displacements as motion-encoder input (ablation)CIDNN · GCADE: 0.021 -> 0.012Improved
Add Gaussian noise to a fraction of training data (ablation)CIDNN · GCMSE: 0%, 5%, 10%, and 20%: 0.0125, 0.0137, 0.0143, and 0.0145Hurt

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