The HiEndo framework revolutionizes gastrointestinal laparoscopy video generation, producing high-resolution, high-fidelity visuals essential for robotic surgery. It leverages a two-stage architecture, first creating low-resolution videos via a diffusion transformer, then enhancing clarity with a super-resolution module. Training on a comprehensive dataset of 61,270 clips, HiEndo outperforms previous models, achieving significant gains in video quality metrics like a 15.1% reduction in Fréchet video distance. This innovation addresses critical needs in surgical visualization.
Journal Article by Wang Z, Zhang Y (…) Zhang Z et 9 al. in Comput Assist Surg (Abingdon)