VLLM data labeler szerző: Aasman Bashyal

The VLLM Data Labeler is a browser add-on that transforms YouTube videos into a powerful data source for AI development. It seamlessly integrates into YouTube watch pages, allowing users to capture specific frames from videos. Each captured frame can then be meticulously annotated with custom labels, providing detailed descriptions and precise locations within the image.
This tool offers flexible export options: users can save individual frames as JPEG images or export them alongside their associated labels in a structured JSON format. This dual export capability is ideal for building comprehensive datasets for training Visual Large Language Models (VLLMs) and other computer vision applications. Features like automatic timestamp-based filenames and persistent local storage ensure an organized and uninterrupted workflow. The intuitive, draggable, and collapsible user interface guarantees a non-intrusive experience, letting you focus on data annotation without disrupting video playback. This add-on is an essential resource for researchers and developers seeking to streamline the creation of high-quality, annotated image datasets from video content.