Research map / Video / Video segmentation and tracking
DisenQ: Disentangling Q-Former for Activity-Biometrics
Shehreen Azad, Yogesh Singh Rawat
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 text encoder supervision to vision encoder (ablation) | DisenQ · NTU RGB-AB | Rank-1: 73.2 -> 77.7 | Improved |
| add DisenQ on top of text encoder (ablation) | DisenQ · NTU RGB-AB | Rank-1: 77.7 -> 82.2 | Improved |
| add DisenQ on top of text encoder (ablation) | DisenQ · Charades-AB | Rank-1: 46.5 -> 49.9 | Improved |
| disentangle biometrics, non-biometrics and motion versus none (ablation) | DisenQ · NTU RGB-AB | Rank-1: 74.2 -> 82.2 | Improved |
| add motion disentanglement to biometrics/non-biometrics split (ablation) | DisenQ · NTU RGB-AB | Rank-1: 76.6 -> 82.2 | Improved |
| add non-biometrics disentanglement to biometrics/motion split (ablation) | DisenQ · NTU RGB-AB | Rank-1: 79.2 -> 82.2 | Improved |
| combine motion with biometrics feature at inference (ablation) | DisenQ · NTU RGB-AB | Rank-1: 80.4 -> 82.2 | Improved |
| use non-biometrics feature for matching (ablation) | DisenQ · NTU RGB-AB | Rank-1: 82.2 -> 3.8 | Hurt |
| use random clothing text instead of VLM non-biometrics descriptions (ablation) | DisenQ · NTU RGB-AB | Rank-1: 9.2% drop | Hurt |
| use three independent Q-Formers instead of unified DisenQ (ablation) | DisenQ · NTU RGB-AB | Rank-1: 0.23% gain with tripled parameters | No clear effect |
| swap EVA-CLIP ViT-G/14 for SigLIP-L (ablation) | DisenQ · NTU RGB-AB | mAP: 43.8 -> 41.6 | Hurt |
| swap LLaVA 1.5 for GPT-4V as VLM (ablation) | DisenQ · NTU RGB-AB | Rank-1: 82.2 -> 82.3 | No clear effect |
Closest papers by meaning
- Activity-Biometrics: Person Identification from Daily Activities (CVPR 2024 · similarity 0.88)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.82)
- Learning Disentangled Identifiers for Action-Customized Text-to-Image Generation (CVPR 2024 · similarity 0.82)
- DIFFER: Disentangling Identity Features via Semantic Cues for Clothes-Changing Person Re-ID (CVPR 2025 · similarity 0.81)
- Person De-reidentification: A Variation-guided Identity Shift Modeling (CVPR 2025 · similarity 0.81)
- Exploring More from Multiple Gait Modalities for Human Identification (AAAI 2025 · similarity 0.81)
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.
