Research map / Image generation / GANs
MG-GAN: A Multi-Generator Model Preventing Out-of-Distribution Samples in Pedestrian Trajectory Prediction
Patrick Dendorfer, Sven Elflein, Laura Leal-Taixé
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 |
|---|---|---|---|
| add multiple generators (4 instead of 1) without classifier or PM-Net (ablation) | single-generator GAN · Synthetic dataset | Precision / Recall: precision 0.46 -> 0.37; recall 0.48 -> 0.68 | Mixed |
| add PM-Net on top of multiple generators (ablation) | multi-generator GAN · Synthetic dataset | Precision: 0.37 -> 0.72 | Improved |
| add classifier C on top of multiple generators (ablation) | multi-generator GAN · Synthetic dataset | Precision: 0.37 -> 0.73 | Improved |
| add PM-Net and classifier together (ablation) | multi-generator GAN · Synthetic dataset | Precision: 0.72 -> 0.77 | Improved |
| use 4 generators instead of one generator at the same total parameter count (ablation) | GAN+L2 · Synthetic dataset | ADE/FDE and recall/precision: 50% (ADE/FDE), 30% (recall/precision) | Improved |
| use expectation sampling with PM-Net instead of random sampling over generators (ablation) | MG-GAN · Synthetic dataset | Recall | Improved |
| vary number of generators n_G from 2 to 8 (ablation) | MG-GAN · ETH/UCY | ADE / FDE: ADE 0.36-0.39, FDE 0.70-0.76 | No clear effect |
| raise lambda_Traj from 0.1 to 5.0 (lambda_Cl = 1.0) (ablation) | MG-GAN (n_G=5) · Synthetic dataset | Precision: 0.64 -> 0.81 | Improved |
| raise lambda_Cl from 0.1 to 5.0 (lambda_Traj = 1.0) (ablation) | MG-GAN (n_G=5) · Synthetic dataset | Precision: 0.73 -> 0.76 | No clear effect |
| vary sigma of PM-Net training from 0.1 to 5.0 (ablation) | MG-GAN (n_G=5) · Synthetic dataset | Precision: 0.77 (0.1), 0.70 (0.5), 0.80 (2.5), 0.72 (5.0) | No clear effect |
| raise generator training samples q from 1 to 20 (l=5) (ablation) | MG-GAN (n_G=5) · Synthetic dataset | Precision: 0.33 -> 0.79 | Improved |
| raise PM-Net samples l from 1 to 5 (q=20) (ablation) | MG-GAN (n_G=5) · Synthetic dataset | Precision: 0.57 -> 0.79 | Improved |
Closest papers by meaning
- Social-BiGAT: Multimodal Trajectory Forecasting using Bicycle-GAN and Graph Attention Networks (NeurIPS 2019 · similarity 0.88)
- A Unified Environmental Network for Pedestrian Trajectory Prediction (AAAI 2024 · similarity 0.86)
- Continuous Trajectory Generation Based on Two-Stage GAN (AAAI 2023 · similarity 0.86)
- Social GAN: Socially Acceptable Trajectories With Generative Adversarial Networks (CVPR 2018 · similarity 0.85)
- TrajCLIP: Pedestrian trajectory prediction method using contrastive learning and idempotent networks (NeurIPS 2024 · similarity 0.85)
- Learning to Generate Diverse Pedestrian Movements from Web Videos with Noisy Labels (ICLR 2025 · similarity 0.83)
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.
