Research map / Detection and segmentation / Open-set and few-shot detection
VLPD: Context-Aware Pedestrian Detection via Vision-Language Semantic Self-Supervision
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 |
|---|---|---|---|
| Initialize visual encoder with CLIP pretrained ResNet-50 (CSP w/ CLIP) (ablation) | CSP baseline · CityPersons Reasonable/Small/HO | MR^-2: 10.96 -> 10.13 (R), 16.05 -> 12.59 (Small), 40.59 -> 38.97 (HO) | Helped |
| Add Vision-Language Semantic (VLS) segmentation with CLIP pseudo labels (ablation) | CSP w/ CLIP · CityPersons Reasonable/Small/HO | MR^-2: 10.13 -> 9.70 (R), 38.97 -> 36.50 (HO); Small unchanged (12.59 -> 12.57) | Helped |
| Add Prototypical Semantic Contrastive (PSC) learning on top of VLS (ablation) | CSP w/ CLIP + VLS · CityPersons Reasonable/Small/HO | MR^-2: 9.70 -> 9.41 (R), 12.57 -> 10.93 (Small), 36.50 -> 34.88 (HO) | Helped |
| Use full CityScapes class set instead of compacted policy for VLS (ablation) | VLS · CityPersons Reasonable/Small/HO | MR^-2: 9.70 -> 10.40 (R), 36.50 -> 37.14 (HO) | Hurt |
| Use 2nd-level CityScapes classes for all classes in VLS (ablation) | VLS · CityPersons Reasonable/Small/HO | MR^-2: 9.70 -> 10.47 (R), 36.50 -> 40.49 (HO) | Hurt |
| Remove human class from VLS class set (ablation) | VLS compacted policy · CityPersons Reasonable | MR^-2: 9.70 -> 10.24 (R) | Hurt |
| Remove traffic sign class from VLS class set (ablation) | VLS compacted policy · CityPersons Reasonable | MR^-2: 9.70 -> 10.11 (R), 36.50 -> 37.22 (HO); Small improves 12.57 -> 12.27 | Hurt |
| Restrict PSC negatives to inner-image instead of cross-image (ablation) | PSC · CityPersons Reasonable/Small/HO | MR^-2: 9.41 -> 10.12 (R), 10.93 -> 12.67 (Small), 34.88 -> 37.90 (HO) | Hurt |
| Take PSC positive prototypes across images instead of inner-image (ablation) | PSC · CityPersons Reasonable/Small/HO | MR^-2: 9.41 -> 10.58 (R), 10.93 -> 13.14 (Small), 34.88 -> 38.08 (HO) | Hurt |
| Concatenate score maps with detection features instead of using E in PSC (ablation) | PSC · CityPersons Reasonable/Small/HO | MR^-2: 9.41 -> 10.04 (R), 10.93 -> 12.93 (Small), 34.88 -> 38.21 (HO) | Hurt |
| Full VLPD (VLS + PSC) vs prior SOTA on Caltech | VLPD · Caltech Reasonable/All/Heavy | MR^-2: Reasonable 2.3 vs PedHunter 2.3 (tie in table; text says 2.27 vs 2.31); Heavy 37.7 vs AP^2M 42.2 | Helped |
| Full VLPD vs prior SOTA on CityPersons (Reasonable, Heavy, Small) | VLPD · CityPersons R/Hea./Small | MR^-2: R 9.4 vs SMPD 9.9; Hea. 43.1 vs 46.4 (MAPD); Small 10.9 vs SML 19.3 | Helped |
Closest papers by meaning
- High-Level Semantic Feature Detection: A New Perspective for Pedestrian Detection (CVPR 2019 · similarity 0.89)
- OmniPT: Unleashing the Potential of Large Vision Language Models for Pedestrian Tracking and Understanding (AAAI 2026 · similarity 0.88)
- A Pedestrian is Worth One Prompt: Towards Language Guidance Person Re-Identification (CVPR 2024 · similarity 0.86)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.85)
- Pedestrian Attribute Recognition: A New Benchmark Dataset and a Large Language Model Augmented Framework (AAAI 2025 · similarity 0.84)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · 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.