Adobes research division, Adobe Research, announced the publication of several scientific papers at the prestigious International Conference on Computer Vision (ICCV 2025) on October 19, 2025. One of the key works was "REGEN: Learning Compact Video Embeddings with a (Re-)Generative Decoder." This research proposes a new method for creating "embeddings"—compressed numerical representations of video that contain its key semantic information. Adobes approach allows these representations to be significantly more compact without losing important data, which drastically reduces computational and memory requirements. The technology paves the way for faster and more efficient AI-powered analysis, search, and editing of video content, which is particularly important for future generations of Adobes creative tools.
Adobe Research Presents REGEN Study on Compact Video Embeddings at ICCV 2025
