Research map / Detection and segmentation / Tracking
MOTSynth: How Can Synthetic Data Help Pedestrian Detection and Tracking?
Matteo Fabbri, Guillem Brasó, Gianluca Maugeri, Orcun Cetintas, Riccardo Gasparini, Aljoša Ošep, Simone Calderara, Laura Leal-Taixé, Rita Cucchiara
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 |
|---|---|---|---|
| train on MOTSynth instead of COCO | YOLOv3 · MOT17 detection | AP: +2.49AP | Improved |
| train on MOTSynth instead of COCO | Faster R-CNN · MOT20 detection | AP: +10.23AP | Improved |
| train on MOTSynth instead of COCO | CenterNet · MOT20 detection | AP: +12.31AP | Improved |
| add more MOTSynth data (subset 1 to subset 4) (ablation) | Faster R-CNN · MOT17 detection | AP: +0.12 | No clear effect |
| add more MOTSynth data (subset 1 to subset 4) (ablation) | YOLOv3 · MOT17 detection | AP: 62.66 -> 71.90 | Improved |
| sample 1/10 of frames instead of 1/60 (fixed image budget) (ablation) | Faster R-CNN · MOTSynth-1 detection (MOT17) | AP: 70.00 -> 76.80 | Improved |
| train on MOTSynth-full instead of VIPER | YOLOv3 · MOT17 detection | AP: +45.25AP | Improved |
| train on MOTSynth-full instead of JTA | Faster R-CNN · MOT17 detection | AP: +18.29 | Improved |
| add people appearance variety (MOTSynth-256 vs JTA, same size) (ablation) | YOLOv3 · MOT17 detection | AP: +9.81AP | Improved |
| fine-tune MOTSynth-trained detector on MOT17 | Faster R-CNN · MOT17 test detection | MODA: 66.7 -> 71.0 | Improved |
| train ReID on MOTSynth split 1 instead of Market1501 | ResNet-50 ReID · MOT17 ReID | mAP: +6.9 | Improved |
| use Market1501 ReID in Tracktor (ablation) | Tracktor · MOT17 train tracking | IDS: 467 -> 179 | Improved |
Closest papers by meaning
- Synthetic Data for Video Surveillance Applications of Computer Vision: A Review (IJCV 2024 · similarity 0.91)
- MixedPeds: Pedestrian Detection in Unannotated Videos Using Synthetically Generated Human-Agents for Training (AAAI 2018 · similarity 0.91)
- Expecting the Unexpected: Training Detectors for Unusual Pedestrians With Adversarial Imposters (CVPR 2017 · similarity 0.90)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.85)
- Learning to Generate Diverse Pedestrian Movements from Web Videos with Noisy Labels (ICLR 2025 · similarity 0.85)
- Multi-View Pedestrian Occupancy Prediction with a Novel Synthetic Dataset (AAAI 2025 · similarity 0.84)
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.
