Research map / Video / Video segmentation and tracking
A Recurrent Encoder-Decoder Network for Sequential Face Alignment
Xi Peng, Rogerio S. Feris, Xiaoyu Wang, Dimitris N. Metaxas
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 |
|---|---|---|---|
| add spatial recurrence with shared weights instead of independent cascaded networks (ablation) | Cascade of three networks · AFLW large pose (>30 deg) | mean error (%): 8.07 -> 7.33 | Improved |
| share weights across steps via spatial recurrence (ablation) | Cascade of three networks · AFLW large pose (>30 deg) | memory (MB): 88.9 -> 29.6 | Improved |
| share weights across steps via spatial recurrence (ablation) | Cascade of three networks · AFLW large pose (>30 deg) | failure rate (%): 25.5 -> 17.3 | Improved |
| add temporal recurrent learning f_tRNN (ablation) | Model without f_tRNN · 300-VW validation, common | mean error (%): 4.52 -> 4.21 | Improved |
| add temporal recurrent learning f_tRNN (ablation) | Model without f_tRNN · 300-VW validation, common | failure rate (%): 3.48 -> 1.71 | Improved |
| add temporal recurrent learning f_tRNN (ablation) | Model without f_tRNN · 300-VW validation, challenging | mean error (%): 6.27 -> 5.64 | Improved |
| add temporal recurrent learning f_tRNN (ablation) | Model without f_tRNN · 300-VW validation, challenging | failure rate (%): 13.3 -> 5.40 | Improved |
| add temporal recurrent learning f_tRNN (ablation) | Model without f_tRNN · 300-VW validation, full | mean error (%): 5.83 -> 5.25 | Improved |
| add supervised identity classification (f_CLS) in the bottleneck (ablation) | Model without f_CLS · 300-VW test, left eye landmark after 10 epochs | validation accuracy: 0.8 -> 0.84 | Improved |
| add supervised identity classification (f_CLS) in the bottleneck (ablation) | Model without f_CLS · 300-VW test accuracy | testing accuracy: approximately 9% improvement | Improved |
| increase spatial recurrent steps to the fourth step (ablation) | 3-step model · AFLW validation | mean error | No clear effect |
| use more spatial recurrent steps on large-pose faces rather than near frontal (ablation) | Spatial recurrent learning · AFLW validation 45-90 deg vs 0-15 deg | mean error improvement: 23.3% improvement on 45-90 deg set and 6.10% improvement on 0-15 deg pose set | Mixed |
Closest papers by meaning
- RED-Net: A Recurrent Encoder–Decoder Network for Video-Based Face Alignment (IJCV 2018 · similarity 0.98)
- Recurrent Convolutional Shape Regression (TPAMI 2018 · similarity 0.89)
- Robust Face Alignment via Deep Progressive Reinitialization and Adaptive Error-driven Learning (TPAMI 2021 · similarity 0.89)
- Two-Stream Transformer Networks for Video-Based Face Alignment (TPAMI 2017 · similarity 0.89)
- Mnemonic Descent Method: A Recurrent Process Applied for End-To-End Face Alignment (CVPR 2016 · similarity 0.88)
- Recursive Spatial Transformer (ReST) for Alignment-Free Face Recognition (ICCV 2017 · similarity 0.88)
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.
