Research map / Detection and segmentation / CNN features
Facial Landmark Detection with Tweaked Convolutional Neural Networks
Yue Wu, Tal Hassner, KangGeon Kim, Gérard G. Medioni, Prem Natarajan
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 tweaking: fine-tune final layers per FC5 cluster | vanilla CNN · validation per-cluster error | error rate: in nearly all clusters, improved by several percents | Improved |
| add tweaking with alignment-sensitive augmentation | vanilla CNN · AFLW | error rate: cut 8% of the error | Improved |
| add tweaking with alignment-sensitive augmentation | vanilla CNN · AFW | error rate: cut 17% of the error | Improved |
| increase number of clusters K up to 64 (ablation) | TCNN · AFW and AFLW | error rate: improves up to K=64, then overfits | Improved |
| increase clusters beyond K=64 (ablation) | TCNN · AFW and AFLW | error rate: overfitting beyond K=64 | Hurt |
| tweak at CL4 instead of FC5 (ablation) | TCNN · AFW and AFLW | error rate: comparable but more tweak time | No clear effect |
| add oversampling and mirroring augmentation (ablation) | TCNN · AFW and AFLW | error rate: proved unsuccessful | Hurt |
| add alignment-sensitive augmentation to single-cluster vanilla CNN (ablation) | vanilla CNN · AFW and AFLW | error rate: failed to show improved results | No clear effect |
| use TCNN landmarks for alignment instead of ERT | recognition pipeline · Janus CS2 (TAR at FAR=1e-2) | true accept rate: .689 -> .718 | Improved |
| use TCNN landmarks for alignment instead of LBF | recognition pipeline · Janus CS2 (TAR at FAR=1e-2) | true accept rate: .643 -> .718 | Improved |
| use TCNN landmarks for alignment instead of CLNF | recognition pipeline · Janus CS2 (TAR at FAR=1e-1) | true accept rate: .897 -> .903 | Improved |
| initialize CLNF with TCNN five points | CLNF · 300W | landmark detection errors | Improved |
Closest papers by meaning
- Semantic Alignment: Finding Semantically Consistent Ground-Truth for Facial Landmark Detection (CVPR 2019 · similarity 0.91)
- Deep Structured Prediction for Facial Landmark Detection (NeurIPS 2019 · similarity 0.91)
- Robust Facial Landmark Detection via a Fully-Convolutional Local-Global Context Network (CVPR 2018 · similarity 0.91)
- Learning Deep Representation for Face Alignment with Auxiliary Attributes (TPAMI 2016 · similarity 0.90)
- Towards Accurate Facial Landmark Detection via Cascaded Transformers (CVPR 2022 · 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 0.90)
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.
