Version 1 focuses on reengineering image towards the saliency objective. The region-of-interest is segmented from the image using SAM2 and hue-rotated based on a predicted angle. While this usually does mark a clear salient region, it is unnatural and may not contextualize the entire salient region, such as the flower above. Needless to say, it is also inefficient. SAM2 is a huge segmentation model that drastically increases time complexity. V1's architecture is too modularized.
InnoColor utilizes a deep-learning guided image transformation to modify visual content, like images, video, and GUI, in both digital software and physical interfaces.
