AI Briefing#vlm
The latest AI and developer news about #vlm, with the original source and a short summary.
The latest AI and developer news about #vlm, with the original source and a short summary.
NVIDIA VFlash reduces latency by up to 5.1x for robot and autonomous driving trajectory prediction.
The framework uses parallel visual primitive drafting to reduce 3D displacement error by up to 61% and accelerate driving inference by up to 49x compared to autoregressive decoding.
Adaptive EVS reduces VLM input tokens by 80% and improves concurrent stream throughput by 46%
Liquid AI has released LFM2.5-VL-3B-DSpark, a draft model for speculative decoding of its VLM LFM2.5-VL-3B, which improves decoding speed by up to 3.13x on the Apple M5 Max.
Based on Qwen3.5-9B, LensVLM maintains original-level accuracy at a 4.3x compression rate and demonstrates performance up to 10.1x.
Meta has released the FLAT model, which integrates understanding and generation by mapping images and text into a single continuous token sequence.
The VisTW benchmark for evaluating VLM performance in Traditional Chinese environments has been released, with Twinkle Eval providing support.
Pinterest built a VLM serving stack using NVIDIA Dynamo and Blackwell GPUs, improving TTFT by 85x.
Inflearn is developing a 'Video Screen Translation' feature that recognizes text within video frames and overlays translated text.
The Qwen team and Huazhong University of Science and Technology (HUST) have released Qwen-Drive-1.0, a vision-language model specialized for autonomous driving, achieving a score of 90.7 on the NAVSIM benchmark when applying RL.
Light Origins has released LightNav-0, a 4B-parameter navigation model that controls multiple robots using only a front-facing RGB camera without depth sensors.
A new mobile agent benchmark named 'MobileWorld' has been released to evaluate capabilities in user interaction and MCP tool utilization.
LG AI Research has released an autonomous vision inspection framework that leverages VLM and Agentic AI to automate the entire process from data labeling to deployment.
An open-source benchmark evaluating 52 text-to-image models with 192 prompts has been released.
Roboflow has launched 'Roboflow Playground,' a tool that allows users to compare and test over 30 of the latest Computer Vision models side by side.
This article introduces architectures and case studies for implementing on-device AI intelligence via external modules without modifying the firmware of existing robots.
MindTopo measured the gap between VLMs' static topological recognition and interaction planning capabilities.
SpyRL implements open-ended LLM self-improvement through game-rule-based rewards.
A study shows that existing metrics for chest X-ray report generation overlook clinical term deletion and bias.
Microsoft has unveiled Mage, a family of efficient multimodal models built at a 4B parameter scale.
The paper proposes the GH-ESD framework, which leverages LLMs and VLMs to discover instance-level error patterns in vision models.
NuMind's open-weight 4B VLM NuExtract3 has been officially released on Ollama, enabling local structured document extraction.
LenVM optimizes LLM inference efficiency and performance by predicting the remaining generation length at the token level.
Summarizes 10 recent AI papers on the themes of agent advancement, security reinforcement, and multi-modality verification.