Research map / 3D vision / Face and shape
Pose-Invariant Face Alignment With a Single CNN
Amin Jourabloo, Mao Ye, Xiaoming Liu, Liu Ren
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 visualization layers during training (ablation) | Proposed single CNN · AFLW | NME (%): 7.18 -> 4.45 | Improved |
| add input image and feature maps (I,F,V) versus (I,V) only (ablation) | Visualization block inputs · AFLW | NME (%): 5.06 -> 4.45 | Improved |
| add input image to (F,V) block inputs (ablation) | Visualization block inputs · AFLW | NME (%): 4.48 -> 4.45 | Improved |
| add mask in visualization layer (ablation) | Visualization layer · AFLW | NME (%): 5.31 -> 4.45 | Improved |
| use more complicated mask 2 instead of original mask (ablation) | Visualization layer · AFLW | NME (%): 4.45 -> 4.49 | No clear effect |
| use 6 blocks with 2 conv layers instead of 4 blocks with 3 conv layers (ablation) | Proposed CNN (12 conv layers) · AFLW | NME (%): 4.61 -> 4.45 | Improved |
| use 6 blocks with 2 conv layers instead of 3 blocks with 4 conv layers (ablation) | Proposed CNN (12 conv layers) · AFLW | NME (%): 4.83 -> 4.45 | Improved |
| add more visualization blocks (sequentially 1 to 6) (ablation) | Proposed · AFLW | NME (%): 9.26 -> 4.45 | Improved |
| add 5% noise to the fully connected parameters of the last block at test time (ablation) | Proposed · AFLW | NME (%): 4.45 -> 5.16 | Hurt |
| add 5% noise to the fully connected parameters of the first block at test time (ablation) | Proposed · AFLW | NME (%): 4.45 -> 4.46 | No clear effect |
| use proposed method instead of LPFA | Large-pose face alignment · AFW | MAPE: 7.43 -> 6.27 | Improved |
| use proposed method instead of 3DDFA+SDM on challenging set | Face alignment · 300W Challenging | NME (%): 9.56 -> 9.88 | Mixed |
Closest papers by meaning
- Large-Pose Face Alignment via CNN-Based Dense 3D Model Fitting (CVPR 2016 · similarity 0.91)
- Face Alignment Across Large Poses: A 3D Solution (CVPR 2016 · similarity 0.90)
- How Far Are We From Solving the 2D & 3D Face Alignment Problem? (And a Dataset of 230,000 3D Facial Landmarks) (ICCV 2017 · similarity 0.89)
- DeCaFA: Deep Convolutional Cascade for Face Alignment in the Wild (ICCV 2019 · similarity 0.89)
- Face Alignment in Full Pose Range: A 3D Total Solution (TPAMI 2017 · similarity 0.89)
- Faster Than Real-Time Facial Alignment: A 3D Spatial Transformer Network Approach in Unconstrained Poses (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.
