Research map / Label-efficient and robust learning / Recognition and long-tail
SynSig2Vec: Learning Representations from Synthetic Dynamic Signatures for Real-World Verification
Songxuan Lai, Lianwen Jin, Luojun Lin, Yecheng Zhu, Huiyun Mao
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 |
|---|---|---|---|
| Optimize average precision of similarity ranking rather than using BCE | SynSig2Vec · MCYT-100 and SVC-Task2, T5 and T1 | EER (%): 2.85 -> 1.71; 7.90 -> 5.50; 6.26 -> 4.65; 14.01 -> 11.96 | Improved |
| Optimize average precision of similarity ranking rather than using triplet loss | SynSig2Vec · MCYT-100 and SVC-Task2, T5 and T1 | EER (%): 2.83 -> 1.71; 8.51 -> 5.50; 6.93 -> 4.65; 15.37 -> 11.96 | Improved |
| Use synthetic signatures in both G1 and G2 instead of genuine signatures and skilled forgeries (ablation) | SynSig2Vec · MCYT-100 T5 and T1 | EER (%): 1.99 -> 1.71; 6.21 -> 5.50 | Improved |
| Use synthetic signatures in both G1 and G2 instead of genuine signatures and skilled forgeries (ablation) | SynSig2Vec · SVC-Task2 T5 and T1 | EER (%): 5.04 -> 4.65; 13.62 -> 11.96 | Improved |
| Add synthetic G1 signatures rather than using genuine signatures for G1 (ablation) | SynSig2Vec with synthetic G2 · MCYT-100 T5 and T1 | EER (%): 1.92 -> 1.71; 6.37 -> 5.50 | Mixed |
| Add synthetic G1 signatures rather than using genuine signatures for G1 (ablation) | SynSig2Vec with synthetic G2 · SVC-Task2 T5 and T1 | EER (%): 4.98 -> 4.65; 13.83 -> 11.96 | Mixed |
| Add synthetic G2 signatures rather than skilled forgeries for G2 (ablation) | SynSig2Vec with synthetic G1 · MCYT-100 T5 and T1 | EER (%): 2.16 -> 1.71; 6.07 -> 5.50 | Mixed |
| Add synthetic G2 signatures rather than skilled forgeries for G2 (ablation) | SynSig2Vec with synthetic G1 · SVC-Task2 T5 and T1 | EER (%): 5.05 -> 4.65; 12.58 -> 11.96 | Mixed |
| Use deeper networks or larger receptive fields (ablation) | SynSig2Vec CNN · Signature verification | EER | No clear effect |
| Add residual connections (ablation) | SynSig2Vec CNN · Signature verification | EER | Hurt |
Closest papers by meaning
- SynSig2Vec: Forgery-Free Learning of Dynamic Signature Representations by Sigma Lognormal-Based Synthesis and 1D CNN (TPAMI 2021 · similarity 0.93)
- A Behavioral Handwriting Model for Static and Dynamic Signature Synthesis (TPAMI 2016 · similarity 0.87)
- Beyond Slow Signs in High-fidelity Model Extraction (NeurIPS 2024 · similarity 0.82)
- Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection (CVPR 2024 · similarity 0.82)
- Training Verification-Friendly Neural Networks via Neuron Behavior Consistency (AAAI 2025 · similarity 0.81)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.81)
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.
