Vision-DeepResearch
What it is
A multimodal deep-research framework and benchmark that enables Large Multimodal Models (LMMs) to perform long-horizon reasoning involving multiple steps and hundreds of search-engine interactions. It includes a comprehensive benchmark (VDR-Bench) and provides datasets for SFT and RL to improve visual and video reasoning.
The problem it solves
Existing LMMs struggle with long-horizon reasoning and complex multi-step visual search tasks.
What you could use it for
- Evaluating long-horizon reasoning capabilities in Multimodal LLMs
- Training models to perform complex visual search and reasoning tasks
- Benchmarking video understanding and reasoning across multiple turns
- Fine-tuning LMMs for deep research tasks involving multi-step reasoning
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.