How to make automatic masking of objects in images

I'm working on a project where I try to detect shirts and change their color. I tried to look for an algorithm in order to implament in the image labeler for masking but so far without any luck, I'm looking for other ways to implament the masking?

Answer 1

[Image Analyst]

It depends on how similar the images are and how much clutter there is in the non-shirt parts of the image. You might try Deep Learning but you'll have to label a bunch of ground truth images, perhaps hundreds of them. If you can't do that then you might need to do it analytically and that will depend on how unique the shirt is in relation to the background. For example, if you have a shirt alone on a white background, that's really easy. If it's in a photo of a bunch of people in a restaurant, then that's a much, much more challenging problem because of multiple shirts at multiple angles, variable lighting, variable contrast with background, various background clutter, etc.

[Trying To Learn]

Do you have any tutorial that is easy to understand for Deep Learning using Ground Truth? I'm doing a research on how to do segmentation and tracing or masking clothes in image with complex background such as selfies.

[Image Analyst]

I do have things like that but nothing yet that is so general purpose as to be a good tutorial. They're all customized somewhat to my setup. But I'm working on it. Maybe some day.