Research map / Label-efficient and robust learning / Recognition and long-tail
A Multi-Level Contextual Model For Person Recognition in Photo Albums
Check 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 iterative photo-level joint inference CRF to the baseline (ablation) | PIPA person recognition · PIPA original, album, time, and day | accuracy: 83.86% -> 84.84%; 78.23% -> 79.13%; 70.29% -> 71.94%; 56.40% -> 57.37% | Helped |
| Add automatically discovered photo-group context to the baseline (ablation) | PIPA person recognition · PIPA original, album, and time | accuracy: 83.86% -> 85.10%; 78.23% -> 79.63%; 70.29% -> 72.02% | Helped |
| Add automatically discovered photo-group context to the baseline (ablation) | PIPA person recognition · PIPA day | accuracy: 56.40% -> 57.33% | Helped |
| Combine person-, photo-, and group-level context (ablation) | PIPA person recognition · PIPA original, album, time, and day | accuracy: 83.86% -> 88.20%; 78.23% -> 83.02%; 70.29% -> 77.04%; 56.40% -> 59.77% | Helped |
| Use album information to form photo groups (ablation) | Multi-level contextual model · PIPA original, album, time, and day | accuracy: 88.20% -> 94.27%; 83.02% -> 83.79%; 77.04% -> 80.64%; 59.77% -> 61.77% | Helped |
| Use identity co-occurrence prior with α = 2 rather than only mutual exclusion with α = 1 (ablation) | Photo-level joint inference · PIPA original | accuracy: 85.74% -> 87.81% | Helped |
| Include identity co-occurrence in photo affinity (ablation) | Photo-group discovery · PIPA | accuracy: around 0.3% on average | Helped |
| Use confidence-aware score fusion rather than concatenating region features (ablation) | Face and body recognition · PIPA original, album, time, and day | accuracy: 71.15% -> 83.86%; 67.81% -> 78.23%; 61.76% -> 70.29%; 54.46% -> 56.40% | Helped |
| Use confidence-aware fusion rather than weighted-average fusion (ablation) | Face and body recognition · PIPA day | accuracy: 52.18% -> 56.40% | Helped |
| Add group-level context to confidence-aware fusion (ablation) | Person recognition · Gallagher Album Exp 1, Exp 2, and Exp 3 | accuracy: 86.34% -> 85.50%; 87.61% -> 86.13%; 85.50% -> 83.19% | Mixed |
| Add photo-level context to confidence-aware fusion (ablation) | Person recognition · Gallagher Album Exp 1, Exp 2, and Exp 3 | accuracy: 86.34% -> 89.08%; 87.61% -> 88.66%; 85.50% -> 87.61% | Helped |
| Add group-level context to the full model (ablation) | Person recognition · Gallagher Album Exp 1, Exp 2, and Exp 3 | accuracy: 88.87%; 86.55%; 86.76% versus photo-level-only 89.08%; 88.66%; 87.61% | Mixed |
Closest papers by meaning
- Unifying Identification and Context Learning for Person Recognition (CVPR 2018 · similarity 0.90)
- Sequential Person Recognition in Photo Albums With a Recurrent Network (CVPR 2017 · similarity 0.89)
- Pose-Aware Person Recognition (CVPR 2017 · similarity 0.87)
- Person De-reidentification: A Variation-guided Identity Shift Modeling (CVPR 2025 · similarity 0.84)
- Learning from Synchronization: Self-Supervised Uncalibrated Multi-View Person Association in Challenging Scenes (CVPR 2025 · similarity 0.83)
- ID-Patch: Robust ID Association for Group Photo Personalization (CVPR 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.