A new framework called SIGMA proposes simulation-in-the-loop collaboration for autonomous driving, using onboard modules for real-time perception while cloud-based VLMs handle high-level reasoning only when…

#OpenSourceAI #AutonomousDriving #VLMs #AIResearch
https://arxiv.org/abs/2610.09520

New arXiv work fine-tunes Google DeepMind's Gemma-4 MoE models for spatial reasoning, showing gains over generalist versions on 2D and 3D rotation tasks. Useful signal that targeted fine-tuning still beats generalist VLMs on structured…

#AI #VLMs #ML #arxiv
https://arxiv.org/abs/2610.04206

ARISE introduces an adaptive agentic framework that improves Vision-Language Models for inflammatory bowel disease imaging by structuring reasoning into a transparent 5-stage workflow with image-grounded self-evaluation. This…

#AIinHealthcare #MedicalImaging #VLMs
https://arxiv.org/abs/2610.04777

How can multimodal AI support low-resource sign languages?

BdSLIG explores Bengali Sign Language instruction generation with VLMs and Sign Parameter-Infused prompting.

CV4A11y @ ICCV 2025

arxiv.org/abs/2508.16076

#Accessibility #BanglaNLP #VLMs

New benchmark OSWorld-Science evaluates VLM-based computer use agents on 146 scientific software tasks spanning molecular drawing, pathology imaging, statistics, and physics simulation.

#AIresearch #VLMs #Benchmarks #AIagents
https://arxiv.org/abs/2609.39903

New arXiv paper benchmarks vision-language models on synapse detection and proofreading in connectomics EM images. Most models perform at chance zero-shot, with only closed and largest open models benefiting from few-shot…

#OpenSourceAI #VLMs #Connectomics
https://arxiv.org/abs/2609.36492