At OmAI Lab, we believe the future of AI extends far beyond pure text. We are dedicated to building the "brains" for next-generation systems by focusing on the intersection of Visual Reasoning and Embodied Agents.
Our research spans across open-vocabulary perception, reinforced vision-language models, and real-time inference. We aim to bridge the critical gap between high-level logical reasoning and fine-grained visual action—building models that don't just "see" the world, but intuitively understand and interact with it.
- 📹 VLX-Flow: A real-time VLM for streaming video understanding .
- 🔍 VLX-Seek: Fine-grained visual perception and grounding for physical AI.
- 🚗 VLX-Go: Efficient general-purpose embodied navigation in the wild.
Models that think, reason, and understand the visual world at a granular level.
- 🌟 VLM-R1: Solving Visual Understanding with Reinforced VLMs. (Highly active)
- 🔎 ZoomEye: Enhancing Multimodal LLMs with human-like zooming capabilities through tree-based image exploration.
- 🌍 ImageRAG: Enhancing ultrahigh-resolution remote sensing imagery analysis.
Foundational spatial understanding optimized for edge and on-premise speeds.
- ⚡ OmDet: Real-time, highly accurate, open-vocabulary end-to-end object detection.
- 🔍 VLM-FO1: Bridging the gap between high-level reasoning and fine-grained perception in Vision-Language Models.
- 📐 GroundVLP: Harnessing Zero-shot Visual Grounding from Vision-Language Pre-training.
Action-oriented intelligence for physical and virtual environments.
- 🛠️ OmAgent: A comprehensive framework to build multimodal language agents for fast prototyping and production.
- 🎯 OmTrackVLA: Open and reproducible research for tracking Vision-Language-Action (VLA) models.
Rigorous standards for the open-source multimodal community.
- 🌍 RS5M: A pioneer work in VLM benchmark for remote sensing.
- 📏 OVDEval: A comprehensive evaluation benchmark for Open-Vocabulary Detection.
- 📝 VL-CheckList: Evaluating vision & language pretraining models with objects, attributes, and relations.