Research map / Video / Video segmentation and tracking
RED-Net: A Recurrent Encoder–Decoder Network for Video-Based Face Alignment
Xi Peng, Rogério Feris, Xiaoyu Wang, Dimitris 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 |
|---|---|---|---|
| use ResNet-based asymmetric encoder-decoder with skip connections instead of VGGNet-based (ablation) | RED-Net · AFLW | mean error (%): 6.85 -> 6.33 | Improved |
| use ResNet-based encoder-decoder instead of VGGNet-based (ablation) | RED-Net · 300-VW | mean error (%): 5.16 -> 4.75 | Improved |
| use ResNet-based encoder-decoder instead of VGGNet-based (ablation) | RED-Net · AFLW | fitting time per image: 43.6ms -> 54.9ms | Hurt |
| two-step recurrent detection-followed-by-regression instead of single-step detection (ablation) | RED-Net · AFLW (challenging) | mean error (%): 8.14 -> 6.97 | Improved |
| two-step recurrent detection-followed-by-regression instead of single-step regression (ablation) | RED-Net · AFLW (challenging) | failure rate (%): 14.5 -> 8.75 | Improved |
| use regression after detection step instead of detection again (ablation) | RED-Net · AFLW (challenging) | failure rate (%): mean error 7.33 -> 6.97; failure 8.20 -> 8.75 | Mixed |
| share weights across steps (recurrent) instead of independent cascade (ablation) | RED-Net · AFLW (challenging) | mean error (%): 6.81 -> 6.33 | Improved |
| share weights across steps (recurrent) instead of independent cascade (ablation) | RED-Net · AFLW (challenging) | memory: 468Mb -> 257Mb | Improved |
| add temporal recurrent learning f_trn (ablation) | RED-Net · 300-VW (full) | failure rate (%): 6.43 -> 2.82 | Improved |
| add temporal recurrent learning f_trn (ablation) | RED-Net · 300-VW (challenging) | mean error (%): 6.27 -> 5.64 | Improved |
| add temporal recurrent learning f_trn (ablation) | RED-Net · 300-VW (common) | mean error (%): 4.52 -> 4.21 | Improved |
| add supervised identity disentangling loss f_cls (ablation) | RED-Net · 300-VW | landmark validation accuracy: 0.8 -> 0.84 at 10 epochs; approximately 9% improvement | Improved |
Later papers that compared with this method
2 later accepted papers report RED-Net* as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- EVDI++: Event-Based Video Deblurring and Interpolation via Self-Supervised Learning (TPAMI 2026 · better on 4 of 4 shared results)
- Separation for Better Integration: Disentangling Edge and Motion in Event-based Deblurring (ICCV 2025 · better on 2 of 2 shared results)
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
- A Recurrent Encoder-Decoder Network for Sequential Face Alignment (ECCV 2016 · similarity 0.98)
- Recurrent Convolutional Shape Regression (TPAMI 2018 · similarity 0.89)
- Two-Stream Transformer Networks for Video-Based Face Alignment (TPAMI 2017 · similarity 0.89)
- Robust Face Alignment via Deep Progressive Reinitialization and Adaptive Error-driven Learning (TPAMI 2021 · similarity 0.88)
- Dynamic Facial Analysis: From Bayesian Filtering to Recurrent Neural Network (CVPR 2017 · similarity 0.88)
- Residual Encoder Decoder Network and Adaptive Prior for Face Parsing (AAAI 2018 · 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.
