Research map / Label-efficient and robust learning / Recognition and long-tail
Partially Shared Multi-Task Convolutional Neural Network With Local Constraint for Face Attribute Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use partially shared networks to combine shared and task-specific representations throughout the network | Face attribute learning · CelebA test | average error rate: 8.75 -> 7.78 | Helped |
| Add identity-based local constraint loss to PS-MCNN | PS-MCNN · CelebA test | average error rate: 7.78 -> 7.02 | Helped |
| Add identity-based local constraint loss to PS-MCNN | PS-MCNN · LFWA | average error rate: 12.88 -> 12.64 | Helped |
| Add shared features through SNet (ablation) | Four independent task networks · CelebA validation | average error rate: 8.84 -> 7.77 | Helped |
| Initialize SNet with face-recognition pretraining and TSNets with attribute-task pretraining (ablation) | PS-MCNN · CelebA validation | average error rate: 7.89 -> 7.51 | Helped |
| Train SNet and TSNets from scratch rather than using face-recognition pretraining for both (ablation) | PS-MCNN · CelebA validation | average error rate: 7.89 -> 7.62 | Helped |
| Remove all task-specific representations and use only SNet (ablation) | PS-MCNN · CelebA validation | average error rate: 7.77 -> 9.07 | Hurt |
| Set SNet channels to 1/8 of TSNet channels (ablation) | PS-MCNN · CelebA validation | average error rate: 7.77 -> 7.91 | Hurt |
| Set SNet channels to 1/2 of TSNet channels (ablation) | PS-MCNN · CelebA validation | average error rate: 7.77 -> 7.81 | No clear effect |
| Set SNet channels equal to TSNet channels (ablation) | PS-MCNN · CelebA validation | average error rate: 7.77 -> 7.82 | No clear effect |
| Connect SNet to TSNets every two layers rather than at every layer (ablation) | PS-MCNN · CelebA validation | average error rate: 7.51 -> 7.80 | Hurt |
| Add partially shared connections instead of using independent task networks (ablation) | Four closely related attribute tasks · Four related attributes | error rate: 4.58 -> 2.99 | Helped |
Closest papers by meaning
- Attributes for Improved Attributes: A Multi-Task Network Utilizing Implicit and Explicit Relationships for Facial Attribute Classification (AAAI 2017 · similarity 0.87)
- Fully-Adaptive Feature Sharing in Multi-Task Networks With Applications in Person Attribute Classification (CVPR 2017 · similarity 0.87)
- A Deep Cascade Network for Unaligned Face Attribute Classification (AAAI 2018 · similarity 0.85)
- Linkage Based Face Clustering via Graph Convolution Network (CVPR 2019 · similarity 0.84)
- Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond (AAAI 2024 · similarity 0.83)
- Joint-Task Regularization for Partially Labeled Multi-Task Learning (CVPR 2024 · similarity 0.82)
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.