#vision-language models

'Overview of leading open-source OCR models and guidance on selecting the right option for printed, handwritten, or multimodal documents.'

'Alibaba Qwen team introduced GUI-Owl and Mobile-Agent-v3, a unified multimodal agent and multi-agent framework that automates GUI tasks across mobile and desktop with state-of-the-art benchmark performance.'

'Liquid AI unveils LFM2-VL, two open-weight vision-language models optimized for fast, low-latency on-device inference, offering 450M and 1.6B variants and easy integration via Hugging Face.'

Mirage introduces a new method for vision-language models to integrate visual reasoning without generating images, significantly enhancing their ability to solve spatial and multimodal tasks.

Google has open-sourced MedGemma 27B Multimodal and MedSigLIP, cutting-edge models designed for scalable multimodal medical reasoning and efficient healthcare AI applications.

Poor product data in fashion leads to lost sales, increased returns, and customer frustration. Multimodal AI offers a scalable solution to improve data accuracy and streamline retail operations.

X-Fusion introduces a dual-tower architecture that adds vision capabilities to frozen large language models, preserving their language skills while improving multimodal performance in image understanding and generation.

Enkrypt AI’s report reveals serious safety flaws in Mistral’s vision-language models that enable generation of harmful content, urging continuous security improvements in multimodal AI systems.

This tutorial covers hands-on implementations of four key vision foundation models—CLIP, DINO v2, SAM, and BLIP-2—highlighting their business applications from product classification to marketing content analysis.

UniME introduces a two-stage framework that significantly improves multimodal representation learning by leveraging textual knowledge distillation and hard negative instruction tuning, outperforming existing models on multiple benchmarks.

A recent study reveals how errors in AI dataset annotations distort the evaluation of vision-language models, advocating for improved human labeling practices to enhance model reliability and reduce hallucinations.

NVIDIA introduces Describe Anything 3B, a multimodal large language model that excels in detailed, region-specific captioning for images and videos, outperforming existing models on multiple benchmarks.