Object recognition is doing a good job, but is still a ways off properly selecting all of the elements of an image or separating multiple parts on a single layer (for an object that may have multiple characters or a single item that is hidden behind others, etc.). It would be nice to see this being enhanced. One other issue along with this is how the magic wand tool and Quick Selection tool also function. In quick selection, when the object colors become exceedingly close, it struggles (and understandably so), but perhaps there is a way to set tolerances or certain color ranges, that could help in this selection process? In both object selection and magic wand, one frustration is always shadows, right? If the AI is trying to recognize an object, perhaps when analyzing a scene it can improve the algorithm for separating a background and recognizing shadows? Or creating a separate layer for the shadow vs. the object and we can make our own adjustments from there? As per another post I just made about a photoshop upscale / enhancement tool improvement to match upscale.media online - this tool would also contribute to improving the selection tool both through object recognition, hard and soft line differentiation. Anyway, on the attached cropped image, I digitized art where the artist struggled to create water ripples and was going to do that in photoshop. You can see from the grass on the bottom (every hue of green), to the reflections of the forest (also hues of green), this is where neither the quick selection or magic wand tool could not differentiate the hues. I probably could have used photoshop to just mirror the image and create the ripples, but I wanted to minimize the photoshopping vs. the original art. However, I did use the upscale.media enhancement tool for kicks just to see what it would do and it clearly delineated the two areas? Anyway - hope this helps!
AdChoices
