Research map / Label-efficient and robust learning / Recognition and long-tail
MOON: A Mixed Objective Optimization Network for the Recognition of Facial Attributes
Ethan M. Rudd, Manuel Günther, Terrance E. Boult
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 one multi-task network on all 40 attributes instead of separate per-attribute networks | MOON (VGG-16) · CelebA test | average classification error (%): 9.78 -> 9.06 | Improved |
| MOON compared with LNets+ANet state of the art | MOON · CelebA test | average classification error (%): 12.70 -> 9.06 | Improved |
| add domain-adaptive re-balancing to a balanced target distribution (ablation) | MOON · CelebAB (balanced error) | average balanced classification error (%): 21.41 -> 12.98 | Improved |
| add domain-adaptive re-balancing (balanced target) evaluated on the unbalanced CelebA test set (ablation) | MOON · CelebA test (unbalanced test set) | average classification error (%): 9.06 -> 13.67 | Hurt |
| test on randomly perturbed (rotation, scale, translation) faces instead of aligned faces | MOON trained on aligned faces · CelebA test | average classification error (%): 9.06 -> 11.62 | Hurt |
| add mis-aligned and mirrored training augmentation (10 copies) on mis-aligned test images (ablation) | MOON · CelebA test (perturbed) | average classification error (%): 11.62 -> 9.50 | Improved |
| add mis-aligned and mirrored training augmentation, evaluated on aligned test images (ablation) | MOON · CelebA test (aligned) | average classification error (%): 9.06 -> 9.23 | Hurt |
| use re-balanced MOON attributes instead of Kumar et al. attributes | attribute-based face verification · LFW verification (View 2) | accuracy (%): 83.43 -> 85.05 | Improved |
| add re-balancing to the MOON network for LFW verification (ablation) | MOON attributes · LFW verification (View 2) | accuracy (%): 84.73 -> 85.05 | No clear effect |
Later papers that compared with this method
3 later accepted papers report MOON as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Pose-guided Enriched Feature Learning for Federated-by-camera Person Re-identification (CVPR 2026 · better on 4 of 4 shared results)
- Transferability and Hardness of Supervised Classification Tasks (ICCV 2019 · better on 1 of 3 shared results)
- Doing the Best We Can With What We Have: Multi-Label Balancing With Selective Learning for Attribute Prediction (AAAI 2018 · better on 1 of 1 shared result)
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
- Doing the Best We Can With What We Have: Multi-Label Balancing With Selective Learning for Attribute Prediction (AAAI 2018 · similarity 0.89)
- Attributes for Improved Attributes: A Multi-Task Network Utilizing Implicit and Explicit Relationships for Facial Attribute Classification (AAAI 2017 · similarity 0.88)
- Attribute-Enhanced Face Recognition With Neural Tensor Fusion Networks (ICCV 2017 · similarity 0.87)
- Heterogeneous Face Attribute Estimation: A Deep Multi-Task Learning Approach (TPAMI 2017 · similarity 0.87)
- Robust RGB-D Face Recognition Using Attribute-Aware Loss (TPAMI 2019 · similarity 0.87)
- Attributes Shape the Embedding Space of Face Recognition Models (ICML 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.
