finetuning with correspondences

[fedeloper]

Can the model be fine-tuned with images and their corresponding labels, without using a depth map?

[TruongKhang]

What do you mean by "corresponding labels" here? semantic segmentation labels?