Research map / Architectures and efficiency / Pruning and efficient tuning
HKAFER: Achieve Visual Parameter-Efficient Fine-Tuning via Heterogeneous Kronecker Adaptation for Facial Expression Recognition
Yu Gao, Haoyu Ji, Zhiyong Wang, Wenze Huang, Qian Dong, Zhihao Yang, Xueting Liu, Weihong Ren, Honghai Liu
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 DBIR router instead of summing adapter outputs (ablation) | HKAFER (DINOv2-ViT-L/14) · RAF-DB | accuracy (%): 92.28 -> 92.70 | Improved |
| use DBIR instead of Semantic Branch only (ablation) | HKAFER (DINOv2-ViT-L/14) · RAF-DB | accuracy (%): 92.44 -> 92.70 | Improved |
| use DBIR instead of Relational Branch only (ablation) | HKAFER (DINOv2-ViT-L/14) · ExpW | accuracy (%): 74.69 -> 74.81 | Improved |
| use heterogeneous Kronecker adapters (HeKA) instead of homogeneous (HoKA) (ablation) | HKAFER (DINOv2-ViT-L/14) · RAF-DB | accuracy (%): 92.24 -> 92.70 | Improved |
| use HeKA instead of LoRA rank 32 (ablation) | HKAFER (DINOv2-ViT-L/14) · RAF-DB | accuracy (%): 91.98 -> 92.70 | Improved |
| use HeKA instead of full fine-tuning (ablation) | DINOv2-ViT-L/14 · RAF-DB | accuracy (%): 89.99 -> 92.70 | Improved |
| use HeKA instead of LoRA rank 32 (ablation) | HKAFER (DINOv2-ViT-L/14) · ExpW | accuracy (%): 74.18 -> 74.81 | Improved |
| set number of adapters N to 4 rather than 1, 2 or 9 (ablation) | HeKA · RAF-DB and ExpW | accuracy (%) | Improved |
| use HKAFER vs CEPrompt under occlusion | CLIP backbone · RAF-DB occlusion | accuracy (%): 90.13 -> 92.25 | Improved |
| use HKAFER vs JADFER in cross-dataset evaluation | CLIP backbone · Cross-dataset mean (R to A, A to R) | accuracy (%): 65.07 -> 70.24 | Improved |
| HKAFER on DINOv2 compared with larger-pretrained baselines on AffectNet and FERPlus | DINOv2-ViT-L/14 · AffectNet | accuracy (%): 68.06 -> 67.66 | Hurt |
Closest papers by meaning
- Personalized Feature Translation for Expression Recognition: An Efficient Source-Free Domain Adaptation Method (ICLR 2026 · similarity 0.88)
- Cross-Domain Facial Expression Recognition: A Unified Evaluation Benchmark and Adversarial Graph Learning (TPAMI 2021 · similarity 0.87)
- TransFER: Learning Relation-Aware Facial Expression Representations With Transformers (ICCV 2021 · similarity 0.87)
- Multi-Energy Guided Image Translation with Stochastic Differential Equations for Near-Infrared Facial Expression Recognition (AAAI 2024 · similarity 0.86)
- Navigating Label Ambiguity for Facial Expression Recognition in the Wild (AAAI 2025 · similarity 0.85)
- Lifting Scheme-Based Implicit Disentanglement of Emotion-Related Facial Dynamics in the Wild (AAAI 2025 · similarity 0.84)
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.
