VisSearch-Bench benchmarks visual search for multimodal agents
Original title: VisSearch-Bench: Evaluating Long Horizon Visual Search in Multimodal Agents
Best model solves only 26.47% of 204 tasks, showing challenges in linking visual evidence.
Multimodal agents use tools for visual search tasks beyond internal knowledge. VisSearch-Bench includes 204 human-written tasks in 7 domains. Across 16 models, even with search tools, complete task success is low, highlighting difficulties in visual context management.
- 7 main domains and 25 subdomains
- Accuracy and Process Score for evaluation
- Best system solves 26.47% tasks
- 5 task types: set-wise, sequential, etc.
- Organizations
- Hugging Face
- AI models
- multimodal agents
From the original article
1 min read at huggingface.co
Multimodal agents use web and image tools to answer questions beyond their internal knowledge. Some search tasks use images mainly to identify an initial entity, then rely on text retrieval. These tasks offer limited insight into how agents connect information across images, objects, and search steps.
By ZhengXian Wu, Qianhao Nie, Zhuohong Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.