Nvidia filed 1 patent in the AI & Machine Learning category.
The filing covers a system that automatically generates training data for computer vision models in gaming applications. The technology crawls game wikis to collect seed images, uses a multimodal language model to label them, and then expands the labeled dataset by processing gameplay footage, eliminating the need for manual annotation.
NVIDIA's single AI and machine learning patent tackles the labor-intensive process of creating labeled training data for game computer vision models. The system begins by automatically gathering seed images from game wikis and related sources, then employs a multimodal large language model to generate labels for these images based on wiki metadata and visual content. Once the seed set is labeled, the technology expands the dataset by identifying and matching similar objects across unlabeled image collections and gameplay video frames. This approach moves beyond traditional methods that relied on manual expert annotation or clustering-based semi-supervised techniques, which the patent notes struggled to distinguish between dynamic game objects, static interface icons, and text elements with sufficient precision.
All data sourced from USPTO patent filings. Google Patents may take several weeks to index recent publications. If a link is unavailable, search for the patent number at USPTO Patent Public Search.