Research map / Label-efficient and robust learning / Recognition and long-tail
A Unified Approach to Kinship Verification
Eran Dahan, Yosi Keller
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 the 1x1-convolution embedding fusion subnetwork instead of concatenation plus FC (ablation) | unified kin verification network · RFIW unrestricted | average accuracy (%): 55.65 -> 79.55 | Improved |
| add the fusion subnetwork on top of raw face embeddings (ablation) | unified kin verification network · RFIW unrestricted | average accuracy (%): 73.65 -> 79.55 | Improved |
| add adaptive family-balanced sampling instead of random sampling (ablation) | unified kin verification network · RFIW unrestricted | average accuracy (%): 75.00 -> 79.55 | Improved |
| add multi-task training across all kinship classes instead of one CNN per class (ablation) | unified kin verification network · RFIW unrestricted | average accuracy (%): 78.42 -> 79.55 | Improved |
| add asymmetric embedding weighting vs no normalization (ablation) | unified kin verification network · RFIW unrestricted | average accuracy (%): 78.90 -> 79.55 | Improved |
| use asymmetric weights instead of symmetric weights (ablation) | unified kin verification network · RFIW unrestricted | average accuracy (%): 78.98 -> 79.55 | Improved |
| vary the BCE loss weight alpha (ablation) | unified kin verification network · RFIW | accuracy | No clear effect |
| switch backbone from VGG-Face to fine-tuned SphereFace | proposed fusion scheme · RFIW FG2018 | average accuracy (%): 71.1 -> 79.6 | Improved |
| add the proposed fusion and classification network on top of SphereFace | SphereFace · RFIW FG2018 | average accuracy (%): 71.2 -> 79.6 | Improved |
| use proposed scheme versus best distance-learning baseline on VGG-Face | VGG-Face · RFIW FG2018 | average accuracy (%): 68.5 -> 71.1 | Improved |
| use proposed scheme versus FG2020 top entry on grandparent-free classes | FG2020 challenge entries · FG2020 (unrestricted) | average accuracy: 0.77 -> 0.78 | Mixed |
Closest papers by meaning
- Kinship Identification through Joint Learning using Kinship Verification Ensembles (ECCV 2020 · similarity 0.90)
- Visual Transformation Aided Contrastive Learning for Video-Based Kinship Verification (ICCV 2017 · similarity 0.88)
- MultiHuman-Testbench: Benchmarking Image Generation for Multiple Humans (NeurIPS 2025 · similarity 0.78)
- PoseBH: Prototypical Multi-Dataset Training Beyond Human Pose Estimation (CVPR 2025 · similarity 0.78)
- SapiensID 2.0: Aligning Human Recognition Foundation Models with Human Perception (NeurIPS 2026 · similarity 0.78)
- SapiensID: Foundation for Human Recognition (CVPR 2025 · similarity 0.78)
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.
