Research map / Label-efficient and robust learning / Recognition and long-tail
Facial Dynamics Interpreter Network: What are the Important Relations between Local Dynamics for Facial Trait Estimation?
Seong Tae Kim, Yong Man Ro
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 locational features (ablation) | Facial dynamics interpreter network · UvA-NEMO age estimation | MAE (years): 4.06 -> 4.00 | Improved |
| add relational importance on top of locational features (ablation) | Facial dynamics interpreter network · UvA-NEMO age estimation | MAE (years): 4.00 -> 3.87 | Improved |
| use relational features of pairs instead of regional-importance aggregation (ablation) | Facial dynamics interpreter network · UvA-NEMO age estimation | MAE (years): 4.26 -> 4.06 | Improved |
| add locational features (ablation) | Facial dynamics interpreter network · UvA-NEMO gender classification | accuracy (%): 88.79 -> 89.35 | Improved |
| add relational importance on top of locational features (ablation) | Facial dynamics interpreter network · UvA-NEMO gender classification | accuracy (%): 89.35 -> 90.08 | Improved |
| use relational features of pairs instead of regional-importance aggregation (ablation) | Facial dynamics interpreter network · UvA-NEMO gender classification | accuracy (%): 88.87 -> 88.79 | No clear effect |
| model local-region relations instead of a holistic LSTM | Proposed method · UvA-NEMO age estimation | MAE (years): 4.02 -> 3.87 | Improved |
| model local-region relations instead of a holistic LSTM | Proposed method · UvA-NEMO gender classification (all) | accuracy (%): 87.10 -> 90.08 | Improved |
| add facial dynamics to image-based CNN | Proposed method vs Image-based CNN · UvA-NEMO gender classification (all) | accuracy (%): 86.94 -> 90.08 | Improved |
| perturb the most important region (right cheek) with a zero vector rather than less important parts (ablation) | Facial dynamics interpreter network · UvA-NEMO age group 37-66 | MAE (years): 7.56 -> 9.00 | Hurt |
| perturb local dynamic features with zero vector (ablation) | Facial dynamics interpreter network · UvA-NEMO age group 37-66 | MAE (years): 5.05 -> 7.56 | Hurt |
| use relational importance for robustness to feature perturbation of left cheek (ablation) | Facial dynamics interpreter network · single 17-year-old sequence | absolute error (years): 7.45 -> 2.38 | Improved |
Closest papers by meaning
- Semantic Relationships Guided Representation Learning for Facial Action Unit Recognition (AAAI 2019 · similarity 0.84)
- From Facial Expression Recognition to Interpersonal Relation Prediction (IJCV 2017 · similarity 0.84)
- Hierarchical Spatio-Temporal Probabilistic Graphical Model with Multiple Feature Fusion for Binary Facial Attribute Classification in Real-World Face Videos (TPAMI 2016 · similarity 0.83)
- Interpreting Face Inference Models Using Hierarchical Network Dissection (IJCV 2022 · similarity 0.83)
- Dynamic Facial Analysis: From Bayesian Filtering to Recurrent Neural Network (CVPR 2017 · similarity 0.83)
- DEEPTalk: Dynamic Emotion Embedding for Probabilistic Speech-Driven 3D Face Animation (AAAI 2025 · similarity 0.82)
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.
