On December 9, Zhiyuan Robot announced an open source VideoDataset. According to reports, VideoDataSet is a high-performance video data loading library developed in depth by the Zhiyuan robot team based on actual AI training requirements. It is packaged based on the NVIDIA Video Codec SDK, and a fully asynchronous assembly line for decoding and training is realized through various means such as multi-decoder scheduling and producer-consumer models, making the decoder utilization rate reach more than 90%.

On December 9, Zhiyuan Robot announced an open source VideoDataset. According to reports, VideoDataSet is a high-performance video data loading library developed in depth by the Zhiyuan robot team based on actual AI training requirements. It is packaged based on the NVIDIA Video Codec SDK, and a fully asynchronous assembly line for decoding and training is realized through various means such as multi-decoder scheduling and producer-consumer models, making the decoder utilization rate reach more than 90%.