Research map / Label-efficient and robust learning / Recognition and long-tail
[Re] Reproducibility Report: Contrastive Learning of Socially-aware Motion Representations
Roopsa Sen, Sidharth Sinha, Animesh Jha, Parv Maheshwari
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 changed | Where | Measured | Effect |
|---|---|---|---|
| tune Social-NCE hyperparameters (tau, horizon, lambda, augmentation) (ablation) | Social-STGCNN · ETH | collision rate (COL): 3.54 -> 3.45 | Improved |
| tune Social-NCE hyperparameters (ablation) | Social-STGCNN · ETH | FDE: 0.678 -> 0.674 | Improved |
| raise contrastive weight lambda from 2 to 16 (ablation) | Social-STGCNN · ETH | tuned loss: 2 -> 16 | Improved |
| add Social-NCE loss | Trajectron++ · ETH/UCY average | collision rate: 35.7% decrease | Improved |
| add Social-NCE loss | Social-STGCNN · ETH/UCY average | collision rate: 35.1% decrease | Improved |
| add Social-NCE loss | imitation learning · robot navigation | collision rate: 68.9% decrease | Improved |
| add Social-NCE loss | trajectory models · ETH/UCY | FDE: less than 1% deviation | No clear effect |
| add Social-NCE loss to Rainbow DQN | reinforcement learning · simulated crowd navigation | episodes to reward 0.6: 4000 -> 2000 episodes | Improved |
| port code to PyTorch Lightning | Social-STGCNN · training time | training time: 8hr 32mins -> 8hr 30mins | No clear effect |
| port code to PyTorch Lightning | Trajectron++ · training time | training time: 4hr 28mins -> 4hr 24mins | No clear effect |
| port code to PyTorch Lightning | Social-STGCNN · ETH | collision rate: 1.11 -> 0.53 | Mixed |
| vary data augmentation hyperparameters (ablation) | Social-NCE · ETH/UCY | accuracy | Mixed |
Closest papers by meaning
- Social NCE: Contrastive Learning of Socially-Aware Motion Representations (ICCV 2021 · similarity 0.86)
- Motion Sensitive Contrastive Learning for Self-Supervised Video Representation (ECCV 2022 · similarity 0.84)
- Learning Long-term Motion Embeddings for Efficient Kinematics Generation (CVPR 2026 · similarity 0.83)
- HuMoCon: Concept Discovery for Human Motion Understanding (CVPR 2025 · similarity 0.83)
- ProMotion: Prototypes As Motion Learners (CVPR 2024 · similarity 0.82)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.82)
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.
