Research map / Architectures and efficiency / Neural architecture search
Searching Central Difference Convolutional Networks for Face Anti-Spoofing
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace vanilla convolution with CDC using θ=0.7 (ablation) | CDCN · OULU-NPU Protocol 1 | ACER (%): 3.8 -> 1.0 | Helped |
| Learn θ separately for each layer instead of fixing it at 0.7 (ablation) | CDCN · OULU-NPU Protocol 1 | ACER (%): 1.0 -> 1.8 | Hurt |
| Allow different searched cells at low, mid, and high levels (ablation) | CDCN++ search · OULU-NPU Protocol 1 | ACER (%): 1.7 -> 1.5 | Helped |
| Select the output node using node attention rather than the last intermediate node (ablation) | CDCN++ search · OULU-NPU Protocol 1 | ACER (%): 1.7 -> 1.4 | Helped |
| Add varied cells and node attention to the search configuration (ablation) | CDCN++ search · OULU-NPU Protocol 1 | ACER (%): 1.7 -> 1.3 | Helped |
| Use a NAS-searched backbone instead of the non-NAS backbone (ablation) | CDCN · OULU-NPU Protocol 1 | ACER (%): 1.0 -> 0.7 | Helped |
| Add MAFM instead of direct multi-level concatenation (ablation) | NAS-searched CDCN backbone · OULU-NPU Protocol 1 | ACER (%): 0.7 -> 0.2 | Helped |
| Use vanilla convolution rather than CDC for MAFM attention (ablation) | CDCN++ · OULU-NPU Protocol 1 | ACER (%): 1.2 -> 0.2 | Helped |
| Set MAFM attention kernels to level-specific 7x7, 5x5, and 3x3 rather than 3x3 at all levels (ablation) | CDCN++ · OULU-NPU Protocol 1 | ACER (%): 0.6 -> 0.2 | Helped |
| Replace level-specific MAFM kernels with 5x5 kernels at every level (ablation) | CDCN++ · OULU-NPU Protocol 1 | ACER (%): 0.2 -> 1.1 | Hurt |
| Replace level-specific MAFM kernels with 3x3 kernels at every level (ablation) | CDCN++ · OULU-NPU Protocol 1 | ACER (%): 0.2 -> 0.6 | Hurt |
Closest papers by meaning
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · similarity 0.86)
- Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · similarity 0.85)
- mmFAS: Multimodal Face Anti-Spoofing Using Multi-Level Alignment and Switch-Attention Fusion (AAAI 2025 · similarity 0.85)
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.85)
- One-Class Face Anti-spoofing via Spoof Cue Map-Guided Feature Learning (CVPR 2024 · similarity 0.85)
- SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining (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.