Why not build a video instance segmentation dataset?

Thanks for your work. The openimages dataset is so huge, while existing video instance segmentation dataset is two small and the quality is not good. Do you have any plan to build a video instance segmentation dataset?