Research map / Video / Video segmentation and tracking
Neural Aggregation Network for Video Face Recognition
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 |
|---|---|---|---|
| Replace average pooling with learned cascaded attention | GoogLeNet-BN face features · IJB-A | TAR@FAR=0.001: 0.771 ± 0.064 -> 0.860 ± 0.012 | Helped |
| Replace average pooling with learned cascaded attention | GoogLeNet-BN face features · IJB-A | TPIR@FPIR=0.01: 0.634 ± 0.109 -> 0.804 ± 0.036 | Helped |
| Add a second, content-aware attention block (ablation) | NAN with single attention · IJB-A | TAR@FAR=0.001: 0.847 -> 0.860 | Helped |
| Add a second, content-aware attention block (ablation) | NAN with single attention · IJB-A | TPIR@FPIR=0.01: 0.778 -> 0.804 | Helped |
| Aggregate images within media before aggregating media features (ablation) | NAN · IJB-A | TPIR@FPIR=0.01: 0.804 ± 0.036 -> 0.817 ± 0.041 | Helped |
| Aggregate images within media before aggregating media features (ablation) | CNN+AvePool · IJB-A | TPIR@FPIR=0.01: 0.634 ± 0.109 -> 0.793 ± 0.044 | Helped |
| Replace average pooling with learned attention aggregation | GoogLeNet-BN face features · YouTube Face | Verification accuracy: 95.20 ± 0.76 -> 95.72 ± 0.64 | Helped |
| Use subject-level rather than video-level aggregation (ablation) | NAN · Celebrity-1000 close-set, 1000 subjects | Rank-1 accuracy: 76.24 -> 77.17 | Helped |
| Use subject-level rather than video-level aggregation (ablation) | CNN+AvePool · Celebrity-1000 close-set, 1000 subjects | Rank-1 accuracy: 74.26 -> 73.41 | Hurt |
| Replace average pooling with NAN subject aggregation | GoogLeNet-BN face features · Celebrity-1000 open-set, 800 subjects | Rank-1 accuracy: 75.12 -> 79.87 | Helped |
Closest papers by meaning
- Deep Layer Aggregation (CVPR 2018 · similarity 0.87)
- Cluster and Aggregate: Face Recognition with Large Probe Set (NeurIPS 2022 · similarity 0.86)
- Predicting Salient Face in Multiple-Face Videos (CVPR 2017 · similarity 0.85)
- FaceCoresetNet: Differentiable Coresets for Face Set Recognition (AAAI 2024 · similarity 0.85)
- Sparsifying Neural Network Connections for Face Recognition (CVPR 2016 · similarity 0.85)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 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.