vision-ai

Google推出Guided Vision功能，集成在Gemini Live中，可通过手机摄像头为盲人或低视力用户提供实时音频描述，帮助识别物体、朗读小字并回答相关追问。

Bonsai-Llama-Jev is an open-source, vision-enabled typed-decision inference system that runs locally with low VRAM and high accuracy, outperforming other systems in a diverse benchmark.

A user discusses the challenges of deploying open models from Western labs due to management restrictions on Chinese models, and seeks recommendations for alternatives in the 120b+ size category with vision capabilities.

User shares configuration for running Vision Qwen 3.8 27B model on a 16GB GPU using beellama, achieving 45 tokens per second decode and 85K context.

Running the DeepSeek-V4-Flash-Vision-Exp 285B MoE model on 10-12x RTX 3090 GPUs achieves over 60-120 tok/s decode speeds with vision and tool support, fully documented for reproducibility.

Roboflow announces that integrating Astra for data labeling in vision AI products reduces time and cost to almost zero, potentially unblocking millions of products. The update highlights advancements where AI models like GPT-6 Astra can easily distinguish between similar visual categories such as sports players.

This article benchmarks various LLMs on their ability to identify mushroom species from images using a large dataset, highlighting accuracy issues and safety concerns for real-world applications.

A lab has open-sourced SenseNova U1.5-Lite, an 8B parameter model for image AI that allows making specific changes to images without destroying other elements, as demonstrated on infographics and posters.

Running the Qwen 3.8 27b model with vision on a single RTX 3090 GPU enables 150k context length, and integration with DeepSeek harness allows for plugin development, though a search engine plugin caused a system crash.

A user thanks the community for assistance in setting up the Qwen 3.8 27B AI model, which now works with HomeAssistant and includes vision capabilities, enabling local AI use.

The post questions whether Qwen 3.5 9B running on low-resource systems can outperform GPT-4o, highlighting its small size of less than 7 GB and claimed superior performance.

NVIDIA discusses three workflows using synthetic data and fine-tuning on its Omniverse and Metropolis platforms to improve vision AI agent accuracy for edge deployment.

AlgoFly AI is a platform for building and deploying vision AI models.

NVIDIA announced new physical AI agent skills at CVPR to accelerate research in autonomous vehicles, robotics, and vision AI, including tools for neural reconstruction, simulation, and reinforcement learning.

A founder shares how a cold DM led to Drizz's first customer, and announces the launch of Drizz, a vision AI agent for mobile app testing, on Product Hunt.