SCAM Benchmark

What is SCAM?

SCAM is the largest and most diverse dataset of real-world typographic attack images to date, containing 1,162 images across hundreds of object categories and…

Released

2025-04-07

Modality

multimodal

Evaluates

多模态, Multimodal, 多模态模型, VLM, 安全对齐, Safety Alignment, 图像理解, Image Understanding, 不支持, Unsupported

Openness

restricted

Importer

OpenCompass Hub

Reported scores

0

Source provenance

• Original evidence https://hub.opencompass.org.cn/dataset-detail/SCAM

SCAM paper, code and dataset

• Paper https://arxiv.org/abs/2504.04893
• Code https://github.com/Bliss-e-V/SCAM
• Dataset https://huggingface.co/datasets/BLISS-e-V/SCAM

Which sources cite SCAM?

• SCAM

Related 多模态, Multimodal, 多模态模型, VLM, 安全对齐, Safety Alignment, 图像理解, Image Understanding, 不支持, Unsupported benchmarks

• 3DCodeBench
• A-Bench
• A-OKVQA
• AbsPyramid
• ActionAtlas
• ActiveFly-Bench
• Ada-LEval
• AECBench