Research map / Label-efficient and robust learning / Recognition and long-tail
Heterogeneous Face Attribute Estimation: A Deep Multi-Task Learning Approach
Hu Han, Anil K. Jain, Wang Sheng Fang, Shiguang Shan, Xilin Chen
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 |
|---|---|---|---|
| train on LFW+ and test on MORPH II instead of same-database training (gender) (ablation) | DMTL · MORPH II | gender accuracy (%): 98.0 -> 89.0 | Hurt |
| train on MORPH II and test on LFW+ instead of same-database training (race) (ablation) | DMTL · LFW+ | race accuracy (%): 94.9 -> 70.5 | Hurt |
| cross-database training on LFWA, test on CelebA (ablation) | DMTL · CelebA (40 attributes) | average accuracy (%): 93.0 -> 70.2 | Hurt |
| cross-database training on CelebA, test on LFWA (ablation) | DMTL · LFWA (40 attributes) | average accuracy (%): 86.0 -> 73.0 | Hurt |
| swap AlexNet shared network for GoogLeNet (ablation) | DMTL · CelebA (40 attributes) | average accuracy (%): 91.98% -> 92.05% | No clear effect |
| joint multi-task learning instead of single-task AlexNet per attribute (ablation) | DMTL · CelebA (eight attributes) | attribute accuracy | Improved |
| pre-train on IMDB-WIKI before fine-tuning on LAPAge2015 (ablation) | DMTL · LAPAge2015 | epsilon-error: 0.449 -> 0.289 | Improved |
| combine LFW+ age and race with LFWA attributes (LFWA+, different annotation sources) (ablation) | DMTL · LFWA+ (42 attributes) | average accuracy of 40 attributes (%): 86.0 -> 83 | Hurt |
| category-specific subnetworks with shared features instead of multi-label regression with same features (Guo and Mu, Yi et al.) | DMTL · MORPH II | age MAE (years): 3.92 -> 3.0 and 3.63 -> 3.0 | Improved |
| AlexNet-size multi-task network versus very deep VGG-16 DEX for age | DMTL · MORPH II | age MAE (years): 3.25 -> 3.0 | Improved |
| joint estimation on small LFWA training set vs MCNN-AUX | DMTL · LFWA (40 attributes) | average accuracy (%): 86.0 vs 86.0 | No clear effect |
Closest papers by meaning
- Attributes for Improved Attributes: A Multi-Task Network Utilizing Implicit and Explicit Relationships for Facial Attribute Classification (AAAI 2017 · similarity 0.89)
- A Survey of Deep Facial Attribute Analysis (IJCV 2020 · similarity 0.88)
- Attributes Shape the Embedding Space of Face Recognition Models (ICML 2025 · similarity 0.86)
- FaceBench: A Multi-View Multi-Level Facial Attribute VQA Dataset for Benchmarking Face Perception MLLMs (CVPR 2025 · similarity 0.85)
- Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond (AAAI 2024 · similarity 0.85)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.85)
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.
