Research map / Label-efficient and robust learning / Recognition and long-tail
Robust RGB-D Face Recognition Using Attribute-Aware Loss
Luo Jiang, Juyong Zhang, Bailin Deng
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 attribute-aware loss to softmax (ablation) | softmax, RGB input · Private RGB-D, train set 1 | Rank-1 identification rate (%): 83.12 -> 92.24 | Improved |
| add attribute-aware loss to softmax + center (ablation) | center loss model, depth input · Private RGB-D, train set 1 | Rank-1 identification rate (%): 88.49 -> 93.29 | Improved |
| add attribute-aware loss to softmax + center (ablation) | center loss model, RGB input · Private RGB-D, train set 1 | Rank-1 identification rate (%): 96.56 -> 96.78 | Improved |
| add depth to RGB input (six-channel) (ablation) | softmax · Private RGB-D, train set 1 | Rank-1 identification rate (%): 83.12 -> 88.93 | Improved |
| train on unbalanced instead of balanced set (ablation) | softmax · Private RGB-D, RGB+Depth | Rank-1 identification rate (%): 88.93 -> 86.59 | Hurt |
| train on unbalanced instead of balanced set (ablation) | softmax + attribute-aware · Private RGB-D, RGB+Depth | Rank-1 identification rate (%): 96.26 -> 96.16 | No clear effect |
| use detected attribute features instead of ground-truth attributes (ablation) | softmax + attribute-aware · Private RGB-D, set 1, RGB+Depth | Rank-1 identification rate (%): 96.26 -> 92.84 | Hurt |
| drop age attribute (use gender + ethnicity) (ablation) | softmax + attribute-aware · Private RGB-D, set 1, RGB+Depth | Rank-1 identification rate (%): 96.26 -> 94.16 | Hurt |
| feature-level fusion at middle instead of signal-level fusion (ablation) | softmax · Private RGB-D, set 1 | Rank-1 identification rate (%): 88.93 -> 89.44 | Improved |
| feature-level fusion at back instead of signal-level fusion (ablation) | softmax · Private RGB-D, set 1 | Rank-1 identification rate (%): 88.93 -> 82.67 | Hurt |
| add attribute-aware loss to softmax (ablation) | softmax, yaw -90 · Small RGB-D pose set | Rank-1 identification rate (%): 62.65 -> 81.15 | Improved |
| add attribute-aware loss to softmax (ablation) | softmax, yaw +90 · Small RGB-D pose set | Rank-1 identification rate (%): 50.12 -> 65.88 | Improved |
Closest papers by meaning
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.88)
- NPT-Loss: Demystifying Face Recognition Losses With Nearest Proxies Triplet (TPAMI 2022 · similarity 0.88)
- Attributes for Improved Attributes: A Multi-Task Network Utilizing Implicit and Explicit Relationships for Facial Attribute Classification (AAAI 2017 · similarity 0.87)
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · similarity 0.87)
- Attributes Shape the Embedding Space of Face Recognition Models (ICML 2025 · similarity 0.87)
- Masked Face Recognition with Generative-to-Discriminative Representations (ICML 2024 · similarity 0.86)
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.
