# (AI^2) FAS meeting

## 2024-09-25

- KeyWords
- Face Anti Spoofing (FAS)
- Currently main method use 'RGB-based', some with single or few images
- CoreML (Align Models to Apple Apps)
- Preprocessing
- Eye Registration
- Post-Processing
- Non-Maximal Supression (NMS)
- MobileViT
- TensorFlow Lite

- Paper
- WACV2023: [DIVT](https://openaccess.thecvf.com/content/WACV2023/html/Liao_Domain_Invariant_Vision_Transformer_Learning_for_Face_Anti-Spoofing_WACV_2023_paper.html)
- Transformer-based
- ICIP 2023: [DGUA_FAS](https://arxiv.org/abs/2310.11758v1?utm_source=uwl.me)
- Cross-domain, New attacks
- CVPR 2024: [One-Class Face Anti-spoofing via Spoof Cue Map-Guided Feature Learning](https://openaccess.thecvf.com/content/CVPR2024/papers/Huang_One-Class_Face_Anti-spoofing_via_Spoof_Cue_Map-Guided_Feature_Learning_CVPR_2024_paper.pdf)

- Goal
- Port to MTK 8390 SoC
- https://www.mediatek.com/iot/products/genio-700
- GPU: Support Tensorflow Lite
- https://mediatek.gitlab.io/aiot/doc/aiot-dev-guide/master/hw/mt8390-soc.html
- 目標 通過 **iBeta** Level 1/2 2. 到達實際商用水準
- https://www.ibeta.com/
- https://www.ibeta.com/iso-30107-3-presentation-attack-detection-confirmation-letters/
- NIST
- Can combine RGB+IR later