InstructFLIP: Exploring Unified Vision-Language Model for Face Anti-spoofing

ACM MM 2025

Kun-Hsiang Lin¹  Yu-Wen Tseng¹  Kang-Yang Huang¹  Jhih-Ciang Wu²  Wen-Huang Cheng¹

¹ National Taiwan University  ² National Taiwan Normal University

   

Overview: InstructFLIP is a unified instruction-tuned framework that leverages vision-language models and a meta-domain strategy to achieve efficient face anti-spoofing generalization without redundant cross-domain training.

Highlights

• This paper proposes InstructFLIP, a novel instruction-tuned VLM framework for FAS, which integrates textual supervision to enhance semantic understanding of spoofing cues.
• We design a content-style decoupling mechanism that explicitly separates spoof-related (content) and spoof-irrelevant (style) information, improving generalization to unseen domains.
• We introduce a meta-domain learning strategy to eliminate training redundancy in cross-domain settings by utilizing diverse image-instruction pairs sampled from a structured meta-domain.
• Experimental results demonstrate that InstructFLIP surpasses SOTA methods across multiple FAS benchmarks, effectively capturing spoof-related patterns through language-guided supervision while substantially reducing training overhead, thereby enhancing its applicability in real-world scenarios.

Instruction for code usage

We recommend using Docker to run the code, which can ensure a consistent environment across different machines.

Clone the repository

Install git-lfs first, then clone the repository:

git clone https://github.com/kunkunlin1221/InstructFLIP.git

Prepare the dataset

see data_preprocess/README.md

Training

1. Build the Docker image

cd docker
bash build.sh
cd ..

2. Train the SOTA InstructFLIP

bash docker/run.sh scripts/train_instruct_flip.sh

Ablations

Loss weights

bash docker/run.sh ablations/loss/$Ablation_Settings.sh # Ablation_Settings: The script name in ablations/loss

Data

bash docker/run.sh ablations/data/$Ablation_Settings.sh # Ablation_Settings: The script name in ablations/data

Branch

bash docker/run.sh ablations/data/train_intruct_flip_$Ablation_Settings.sh # Ablation_Settings: The script name in ablations/branch

Replace llm with multi-classes heads

bash docker/run.sh ablations/data/train_mc.sh

Citation

If you're using this work in your research or applications, please cite using this BibTeX:

@InProceedings{lin2025instructflip,
  author    = {Kun-Hsiang Lin and Yu-Wen Tseng and Kang-Yang Huang and Jhih-Ciang Wu and Wen-Huang Cheng},
  title     = {InstructFLIP: Exploring Unified Vision-Language Model for Face Anti-spoofing},
  booktitle = {Proceedings of the 33rd ACM International Conference on Multimedia},
  year      = {2025},
  organization = {ACM},
}