Research map / Detection and segmentation / Open-set and few-shot detection
Artificial Dummies for Urban Dataset Augmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 75 DummyNet samples to 100 real samples | Person classifier · Cityscapes low-data, 1% FPR | miss rate: 0.980 -> 0.763 | Helped |
| Add 75 DummyNet samples to 100 real samples | Person classifier · Cityscapes low-data, 10% FPR | miss rate: 0.682 -> 0.425 | Helped |
| Increase synthetic sample count beyond 75 in the Cityscapes low-data experiment | DummyNet-augmented person classifier · Cityscapes low-data | miss rate: 0.763 -> 0.836 -> 0.861 -> 0.790 | Mixed |
| Add DummyNet-generated night samples to real training data | Person classifier · NightOwls classification, 12,000 real samples, 10% FPR | miss rate: 0.28 -> 0.22 | Helped |
| Augment CityPersons with DummyNet samples | CSP · CityPersons reasonable | LAMR: 11.44% -> 10.25% | Helped |
| Augment day-time training data with DummyNet samples and night backgrounds | Faster R-CNN ResNet-50 · NightOwls detection reasonable | LAMR: 41.90% -> 24.95% | Helped |
| Vary pedestrian pose instead of fixing it (ablation) | DummyNet · NightOwls ablation, 0 real samples, 1% FPR | miss rate: 0.891 -> 0.706 | Helped |
| Use a learned person mask instead of a keypoint convex hull (ablation) | DummyNet · NightOwls ablation, 0 real samples, 1% FPR | miss rate: 0.770 -> 0.706 | Helped |
| Restrict appearance conditioning to darker input people rather than arbitrary appearances (ablation) | DummyNet · NightOwls ablation, 0 real samples, 10% FPR | miss rate: 0.524 -> 0.461 | Helped |
| Condition appearance on input person images rather than Gaussian latent samples (ablation) | DummyNet · NightOwls ablation, 0 real samples, 1% FPR | miss rate: 0.770 -> 0.706 | Helped |
| Vary appearance instead of freezing it to one sample (ablation) | DummyNet · NightOwls ablation, 0 real samples, 1% FPR | miss rate: 0.875 -> 0.706 | Helped |
| Condition generation on the target background instead of using one background followed by post-processing (ablation) | DummyNet · NightOwls ablation, 0 real samples, 1% FPR | miss rate: 0.877 -> 0.706 | Helped |
Closest papers by meaning
- CityPersons: A Diverse Dataset for Pedestrian Detection (CVPR 2017 · similarity 0.87)
- Generalizable Pedestrian Detection: The Elephant in the Room (CVPR 2021 · similarity 0.87)
- Expecting the Unexpected: Training Detectors for Unusual Pedestrians With Adversarial Imposters (CVPR 2017 · similarity 0.87)
- How Far Are We From Solving Pedestrian Detection? (CVPR 2016 · similarity 0.86)
- Pedestrian-Centric 3D Pre-collision Pose and Shape Estimation from Dashcam Perspective (NeurIPS 2024 · similarity 0.83)
- Self-Supervised Likelihood Estimation with Energy Guidance for Anomaly Segmentation in Urban Scenes (AAAI 2024 · 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.