Google has released a dataset of links to 8 million YouTube videos called YouTube-8M to help advance machine learning research. Google annotated each video in YouTube-8M by their subject matter, using 4,800 different labels, which can serve as useful metadata for video interpretation programs. The videos in YouTube-8M makeup over 500,000 hours of video, which is substantially larger than a previous dataset Google released called Sports-1M, which contained one million videos, and likely the largest annotated video dataset available to researchers.
Helping Computers Understand Videos
previous post