Research map / Architectures and efficiency / CNN design
Optimizing Filter Size in Convolutional Neural Networks for Facial Action Unit Recognition
Shizhong Han, Zibo Meng, Zhiyuan Li, James O'Reilly, Jie Cai, Xiaofeng Wang, Yan Tong
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 |
|---|---|---|---|
| learn filter sizes in all three convolutional layers instead of fixing them | CNN on BP4D · BP4D development | F1: 0.499 -> 0.515 | Improved |
| learn filter sizes in all three convolutional layers instead of fixing them | CNN on BP4D · BP4D development | 2AFC: 0.664 -> 0.691 | Improved |
| learn the filter size only in the first convolutional layer rather than all three (ablation) | OFS-CNN on BP4D · BP4D development | F1: 0.515 -> 0.501 | Hurt |
| learn the filter size only in the first convolutional layer rather than all three (ablation) | OFS-CNN on BP4D · BP4D development | 2AFC: 0.666 -> 0.691 | Hurt |
| optimize filter sizes rather than use fixed 5×5 filters | CNN on BP4D at 64×48 · BP4D development, 64×48 | F1: 0.499 -> 0.515 | Improved |
| optimize filter sizes rather than use fixed 5×5 filters | CNN on BP4D at 128×96 · BP4D development, 128×96 | F1: 0.522 -> 0.533 | Improved |
| optimize filter sizes rather than use fixed 5×5 filters | CNN on BP4D at 256×192 · BP4D development, 256×192 | F1: 0.478 -> 0.537 | Improved |
| replace fixed filter sizes with learned filter sizes | CNN on DISFA · DISFA | F1: 0.515 -> 0.553 | Improved |
| replace fixed filter sizes with learned filter sizes | CNN on DISFA · DISFA | 2AFC: 0.850 -> 0.850 | No clear effect |
| use OFS-CNN instead of GoogLeNet's multiple-filter-size architecture | AU recognition on BP4D · BP4D | F1: 0.531 -> 0.537 | Improved |
| use OFS-CNN instead of GoogLeNet | AU recognition on BP4D · BP4D | testing speed: more than 6 times faster | Improved |
Closest papers by meaning
- Deep Facial Action Unit Recognition From Partially Labeled Data (ICCV 2017 · similarity 0.88)
- Incremental Boosting Convolutional Neural Network for Facial Action Unit Recognition (NeurIPS 2016 · similarity 0.88)
- Deep Region and Multi-Label Learning for Facial Action Unit Detection (CVPR 2016 · similarity 0.87)
- Action Unit Detection With Region Adaptation, Multi-Labeling Learning and Optimal Temporal Fusing (CVPR 2017 · similarity 0.86)
- Weakly-Supervised Deep Convolutional Neural Network Learning for Facial Action Unit Intensity Estimation (CVPR 2018 · similarity 0.86)
- Pre-training Strategies and Datasets for Facial Representation Learning (ECCV 2022 · 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.
