Closed jianweif closed 2 years ago
Good question! You can annotate the mask your self. Just make sure that the masks are always bigger than the object (i.e. contain the whole object), and don't contain much background pixels. Our model should be robust enough to handle it.
The masks in provided data seem perfect. I am working on my own data and wondering if it can work with noisy foreground mask coming from a saliency detection model or it has to be perfect foreground mask. Thanks!