Research map / Label-efficient and robust learning / Recognition and long-tail
SPL-Net: Spatial-Semantic Patch Learning Network for Facial Attribute Recognition with Limited Labeled Data
Yan Yan, Ying Shu, Si Chen, Jing‐Hao Xue, Chunhua Shen, Hanzi Wang
This paper tried a common change: Add a reconstruction / auxiliary loss
add all three auxiliary tasks (two-stage pretraining) to baseline · CelebA 0.02% labeled · accuracy (%): 76.34 -> 79.33 Improved
What 987 accepted papers report for this change
- 882 improved
- 41 no clear effect
- 20 mixed
- 44 hurt
- OPTION: Optimal Transport–Guided Flow Matching for Incomplete and Unaligned Multi-View Clustering ICML 2026 · add reconstruction loss L_rec · Scene-15 (aligned, 70% missing) · ACC: 32.4 -> 37.5Improved
- Detached Skip-Links and $R$-Probe: Decoupling Feature Aggregation from Gradient Propagation for MLLM OCR ICML 2026 · add R-probe auxiliary loss · OCRBench · score: 714.0 -> 721.0Improved
- Rethinking Time-Series Imputation as Conditional Inference along Temporal Evolution ICML 2026 · add reconstruction loss on observed entries (joint vs missing-only supervision) · ETTh1 10% missing · Masked MAE: 0.2002 -> 0.1658Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 all three auxiliary tasks (two-stage pretraining) to baseline (ablation) | baseline PreAct ResNet-18 · CelebA 0.02% labeled | accuracy (%): 76.34 -> 79.33 | Improved |
| add all three auxiliary tasks to baseline (ablation) | baseline · LFWA 0.5% labeled | accuracy (%): 4.51% improvement | Improved |
| add all three auxiliary tasks to baseline (ablation) | baseline · MAAD 0.02% labeled | accuracy (%): 63.04 -> 69.01 | Improved |
| add multi-branch shared subnetwork (MSS) to baseline (ablation) | baseline · CelebA/LFWA/MAAD low label | accuracy (%): 0.58% on CelebA, 0.98% on LFWA, 0.57% on MAAD | Improved |
| keep whole region branch rather than feature fusion block (ablation) | SPL-Net · CelebA 0.02% labeled | accuracy (%): 78.21 -> 79.33 | Improved |
| add spatial mutual exclusion (SME) loss (ablation) | SPL-Net · CelebA 0.02% labeled | accuracy (%): 78.83 -> 79.33 | Improved |
| add adversarial training in MSS (ablation) | SPL-Net · CelebA 0.02% labeled | accuracy (%): 78.89 -> 79.33 | Improved |
| use adversarial training instead of L2 contrastive loss (ablation) | SPL-Net · CelebA 0.02% labeled | accuracy (%): 78.01 -> 79.33 | Improved |
| use two-stage learning instead of one-stage semi-supervised (ablation) | SPL-Net · CelebA 0.02% labeled | accuracy (%): 75.23 -> 79.33 | Improved |
| use SE-based DCA attention instead of CBAM blocks (ablation) | SPL-Net · CelebA 0.02% labeled | accuracy (%): 79.03 -> 79.33 | Improved |
| add patch classification task to PRT+PST (ablation) | SPL RS · CelebA/LFWA/MAAD low label | accuracy (%): 0.37%, 1.15%, 0.62% | Improved |
| add patch segmentation task to PRT+PCT (ablation) | SPL RC · CelebA/LFWA/MAAD low label | accuracy (%): 0.83%, 1.57%, 1.25% | Improved |
Closest papers by meaning
- Learning Spatial-Semantic Relationship for Facial Attribute Recognition With Limited Labeled Data (CVPR 2021 · similarity 0.93)
- Improving Facial Attribute Prediction Using Semantic Segmentation (CVPR 2017 · similarity 0.85)
- On Symbiosis of Attribute Prediction and Semantic Segmentation (TPAMI 2019 · similarity 0.83)
- Attributes for Improved Attributes: A Multi-Task Network Utilizing Implicit and Explicit Relationships for Facial Attribute Classification (AAAI 2017 · similarity 0.83)
- Self-Supervised Facial Representation Learning with Facial Region Awareness (CVPR 2024 · similarity 0.81)
- Attributes Shape the Embedding Space of Face Recognition Models (ICML 2025 · similarity 0.80)
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.
