Is there a way to train Adobe Sensei to get more accurate background cutouts for a specific object?

[matthewm73187289]

Using the quick action background removal tool I have been separating car images from their background. All the car images that I have are almost the exact same positioning, framing, etc. I would like to batch process this with a photshop action, but sometimes the cutouts aren't accurate enough because of shadows or certain details. Is there a way that I can train Sensei with a car dataset in Photoshop to recogniz the car for a better cutout?

[Mylenium]

The dataset would have to be thousands of images and in fact would have to feature much more varied shots. KI doesn't work like a static selection as you seem to assume. In order to understand what actually is a shadow is, the KI needs to understand the "flow" of the image, not its absolute pixel data. So unless you literally have hundreds and thousands of images there is no way to consciously train any KI and you have to wait until the day the server-side databases have accumulated enough samples from users all across the world. What you currently achieve is probably as good as it gets for the time being.

Mylenium

[matthewm73187289]

I do have a dataset of 16,000 cars that have varied shots that I could use. The ones I need to cutout are in the same positioning and framing though. You said I would have to wait for the server-side database to accumulate enough images for the process to get better meaning I can't alter in any way the selection tool to refelect the dataset I have on a local level?

[Derek Cross]

Presumably, you are using the Select > Subject feature?

[matthewm73187289]

Essentially yes, the quick action is the remove background button on quick actions under properties. It selects the subject and deletes the inverse.