Research map / Video / Video segmentation and tracking
Deep Spatial Gradient and Temporal Depth Learning for Face Anti-Spoofing
Zezheng Wang, Zitong Yu, Chenxu Zhao, Xiangyu Zhu, Yunxiao Qin, Qiusheng Zhou, Feng Zhou, Zhen Lei
Check 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 contrastive depth loss to the raw backbone (ablation) | Face anti-spoofing depth model · OULU-NPU Protocol 3 | ACER (%): 6.25±3.20 -> 5.07±1.83 | Improved |
| Replace vanilla convolutions with RSGB (ablation) | Model 2 · OULU-NPU Protocol 3 | ACER (%): 5.07±1.83 -> 3.19±0.90 | Improved |
| Add STSTB (ablation) | RSGB depth model · OULU-NPU Protocol 3 | ACER (%): 3.19±0.90 -> 2.99±0.72 | Improved |
| Add ConvGRU (ablation) | RSGB + STSTB model · OULU-NPU Protocol 3 | ACER (%): 2.99±0.72 -> 2.85±0.49 | Improved |
| Add binary supervision (ablation) | Model 5 · OULU-NPU Protocol 3 | ACER (%): 2.85±0.49 -> 2.71±0.63 | Improved |
| Replace STSTB with OFF (ablation) | Temporal depth model · OULU-NPU Protocol 3 | ACER (%): 2.71±0.63 -> 3.20±1.00 | Hurt |
| Use actual rather than generated depth supervision (ablation) | Model 7 · DMAD | ACER (%): 6.33 -> 4.55 | Improved |
| Set the temporal refinement weight α to 0.6 rather than using only spatial depth (α = 0.0) (ablation) | Spatio-temporal depth refinement · OULU-NPU Protocol 3 | ACER (%) | Improved |
| Vary frame sampling interval from 1 to 3 frames (ablation) | Spatio-temporal architecture · OULU-NPU Protocol 3 sub-protocol | ACER (%): 3.347% -> 2.927% | Improved |
| Increase frame sampling interval from 3 to 5 frames (ablation) | Spatio-temporal architecture · OULU-NPU Protocol 3 sub-protocol | ACER (%): 2.927% -> 4.223% | Hurt |
| Increase frame sampling interval from 3 to 7 frames (ablation) | Spatio-temporal architecture · OULU-NPU Protocol 3 sub-protocol | ACER (%): 2.927% -> 2.934% | No clear effect |
| Use the proposed method instead of Auxiliary | Face anti-spoofing model · SiW → OULU-NPU Protocol 3 | ACER (%): 13.8±5.7 -> 14.6±4.8 | Hurt |
Closest papers by meaning
- Learning Deep Models for Face Anti-Spoofing: Binary or Auxiliary Supervision (CVPR 2018 · similarity 0.92)
- Deep Learning for Face Anti-Spoofing: A Survey (TPAMI 2022 · similarity 0.92)
- Face Anti-Spoofing: Model Matters, so Does Data (CVPR 2019 · similarity 0.90)
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.88)
- SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining (AAAI 2025 · similarity 0.87)
- One-Class Face Anti-spoofing via Spoof Cue Map-Guided Feature Learning (CVPR 2024 · similarity 0.87)
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.
