Research map / Label-efficient and robust learning / Recognition and long-tail
Generate to Adapt: Resolution Adaption Network for Surveillance Face Recognition
Han Fang, Weihong Deng, Yaoyao Zhong, Jiani Hu
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 feature adaption to MR-GAN synthesized LR faces (ablation) | MR-GAN · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.1736 -> 0.2287 | Improved |
| add feature adaption to Down-Sampling LR faces (ablation) | Down-Sampling · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.1486 -> 0.1890 | Improved |
| add feature adaption to High-to-Low LR faces (ablation) | High-to-Low · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.0916 -> 0.2132 | Improved |
| add feature adaption to Cycle-GAN LR faces (ablation) | Cycle-GAN · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.1399 -> 0.2058 | Improved |
| replace Down-Sampling with MR-GAN synthesis (ablation) | ResNet34 fine-tuning · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.1486 -> 0.1736 | Improved |
| replace Cycle-GAN with MR-GAN synthesis (ablation) | ResNet34 fine-tuning · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.1399 -> 0.1736 | Improved |
| replace High-to-Low with MR-GAN synthesis (ablation) | ResNet34 fine-tuning · IJB-C TinyFace TAR@FAR 1e-4 | TAR: 0.0916 -> 0.1736 | Improved |
| add translator in translation gate (ablation) | RAN · LFW | accuracy: 97.7 -> 98.7 | Improved |
| add MR feature adaption on large-scale MS-Celeb-1M (ablation) | HR model · IJB-C TinyFace TAR@FAR 1e-4 (MS-Celeb-1M) | TAR: 0.1801 -> 0.2635 | Improved |
| train on MR-GAN LR data instead of HR data (large-scale) (ablation) | ResNet50 ArcFace · IJB-C TinyFace TAR@FAR 1e-2 (MS-Celeb-1M) | TAR: 0.5591 -> 0.5025 | Hurt |
| add RAN to ArcFace without fine-tuning | ArcFace · SCface d1 | rank-1 accuracy: 35.00 -> 70.50 | Improved |
| use RAN instead of ArcFace on QMUL-SurvFace | ArcFace · QMUL-SurvFace TPIR20@FPIR 10% | TPIR20: 10.1 -> 14.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 3 of 3 shared results
- SphereFace — SphereFace: Deep Hypersphere Embedding for Face Recognition (CVPR 2017) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Finding Tiny Faces in the Wild With Generative Adversarial Network (CVPR 2018 · similarity 0.86)
- Face Reconstruction from Facial Templates by Learning Latent Space of a Generator Network (NeurIPS 2023 · similarity 0.85)
- 3D-Aided Dual-Agent GANs for Unconstrained Face Recognition (TPAMI 2018 · similarity 0.85)
- Dual-Agent GANs for Photorealistic and Identity Preserving Profile Face Synthesis (NeurIPS 2017 · similarity 0.84)
- GAN Prior Embedded Network for Blind Face Restoration in the Wild (CVPR 2021 · similarity 0.84)
- Teaching Where to Look: Attention Similarity Knowledge Distillation for Low Resolution Face Recognition (ECCV 2022 · 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.
