新品發表｜主題 | LAETimes 低空日報

Technology

NVIDIA Cosmos 3 Post-Training Completed in One Day: AI Coding Agents Boost Visual Reasoning Model Accuracy Beyond 90%

Autonomous coding AI agents are transforming visual reasoning model development by automating data formatting, container setup, training scripts, and hyperparameter tuning — tasks that traditionally take days. Applied to NVIDIA Cosmos 3, the entire post-training pipeline can now be completed in a single day with minimal human intervention, pushing model accuracy above 90%. The breakthrough holds particular significance for drone and autonomous vehicle applications that rely on high-precision computer vision.

AI自動駕駛電腦視覺

about 1 month ago

Source: NVIDIA Robotics Blog

新品發表

Joby Aviation's Electric Skies Tour to Debut at EAA AirVenture Oshkosh

France Integrates 68mm Laser-Guided Rockets on Rafale, Joining F-16 and Eurofighter Typhoon in Counter-Drone Role

BAE Systems Completes Integration Testing of BATS Layered Counter-Drone System, Reaching Eight-Month Development Milestone

US Air Force Seeks Dronebuster Handheld C-UAS Devices for Space Force Base

Titan®-MS C-UAS Awarded $80.5M Contract to Protect U.S. Air Force Global Strike Command Bases

Swedish Startup Unveils 'Radio-Silent' Counter-Drone Interceptor System

Farnborough Airshow: Eve Air Mobility Showcases Full-Scale Mock-Up and Flight Test Progress

NVIDIA Cosmos 3 Post-Training Completed in One Day: AI Coding Agents Boost Visual Reasoning Model Accuracy Beyond 90%

Leaderboard Lessons: What 5,000 Kaggle Competitors Taught Us About Improving AI Reasoning

New Balance Niobium Concept 1: A 3-in-1 Hiking Boot That Unzips Into a Camp Slipper

World's First Airworthiness-Certified Quantum Navigation Backup System Debuts, Enabling GPS-Free Flight

SkyDrive Completes Simulated Tourism Flight Demo in Yamaguchi Prefecture, Reaching 86 km/h