Research map / Detection and segmentation / Open-set and few-shot detection
Spoof Trace Disentanglement for Generic Face Anti-Spoofing
Yaojie Liu, Xiaoming 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 |
|---|---|---|---|
| add the trace GAN decoder (step 2) with hierarchical traces on top of step 1 (ablation) | PhySTD · SiW-M Protocol I overall | EER (%): 4.3 -> 2.8 | Improved |
| add synthesized spoof and harder-sample training (step 3) (ablation) | PhySTD · SiW-M Protocol I overall | EER (%): 2.8 -> 2.5 | Improved |
| add synthesized spoof and harder-sample training (step 3) (ablation) | PhySTD · SiW-M Protocol I overall | ACER (%): 4.6 -> 2.8 | Improved |
| add synthesized spoof and harder-sample training (step 3) (ablation) | PhySTD · SiW-M Protocol I overall | TPR@FNR=0.5% (%): 89.7 -> 91.2 | Improved |
| use hierarchical multi-component traces B,C,T instead of a single trace (ablation) | PhySTD Step1+Step2 · SiW-M Protocol I overall | EER (%): 5.8 -> 2.8 | Improved |
| use hierarchical multi-component traces B,C,T instead of a single trace (ablation) | PhySTD Step1+Step2 · SiW-M Protocol I overall | ACER (%): 13.8 -> 4.6 | Improved |
| add trace GAN decoder (step 2) to step 1 (ablation) | PhySTD · SiW-M Protocol I overall | TPR@FNR=0.5% (%): 59.3 -> 89.7 | Improved |
| add image decomposition to the auxiliary baseline (step 1) (ablation) | Auxiliary FAS · SiW-M Protocol I overall | EER vs ACER (%): EER 6.7 -> 4.3; ACER 6.3 -> 10.9 | Mixed |
| classify spoof medium on estimated traces instead of original images | PhySTD traces with AlexNet · Oulu-NPU Protocol 1 5-class spoof medium classification | accuracy (%): 80.6 -> 83.3 | Improved |
| extend ECCV preliminary version with SGF, hierarchical supervision and open-set design | PhySTD · SiW-M Protocol I overall | EER (relative reduction): 47.9% relative EER reduction; TDR improved by 29.5% | Improved |
| extend ECCV preliminary version to full PhySTD | PhySTD · OULU-NPU Protocols 1 and 3 | ACER (relative reduction): 63.6% and 32.1% | Improved |
| apply PhySTD to unknown attacks compared with preliminary version | PhySTD · SiW-M Protocol II average | EER, ACER, TPR (relative): 11.7% EER, 19.0% ACER, 50.4% TPR | Improved |
Closest papers by meaning
- On Disentangling Spoof Trace for Generic Face Anti-Spoofing (ECCV 2020 · similarity 0.96)
- Learning Polysemantic Spoof Trace: A Multi-Modal Disentanglement Network for Face Anti-spoofing (AAAI 2023 · similarity 0.90)
- Feature Generation and Hypothesis Verification for Reliable Face Anti-spoofing (AAAI 2022 · similarity 0.89)
- One-Class Face Anti-spoofing via Spoof Cue Map-Guided Feature Learning (CVPR 2024 · similarity 0.87)
- SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining (AAAI 2025 · similarity 0.86)
- Rethinking Generalizable Face Anti-spoofing via Hierarchical Prototype-guided Distribution Refinement in Hyperbolic Space (CVPR 2024 · similarity 0.85)
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.
