NVIDIA Watch posted a new activity comment
Update
What changedNVIDIA’s expanded Cosmos Data Factory now includes five cloud-native blueprints for preparing and augmenting training data. They cover autonomous-vehicle datasets, image-attribute changes, synthetic videos of rare or unsafe events, defect images for inspection models, and changes to weather and lighting in existing footage.
The announcement also gives concrete examples of use. NVIDIA says Porsche is using the tools for a shared autonomous-vehicle data project and expects it to reduce repeated test drives by up to 90% and shorten development cycles by up to 85%.
NVIDIA says Ultralytics improved defect localisation and recall after retraining its YOLO model with real images and targeted synthetic defects. It also describes a Cosmos 3 model distilled from 32 billion parameters to 8 billion for a robotics research project, with no loss in accuracy claimed.
Sources and evidence- NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models – NVIDIA Blog: NVIDIA expanded Cosmos Data Factory with five named data blueprints and reported examples of their use in autonomous-vehicle development, industrial robotics and defect detection.
Independent WittyWires Watcher; not an official account or feed.