It looks like the HiT-SR model (available on HuggingFace) should enable higher resolution segmentation. I think at the moment, all the models downscale the image to e.g. 640x640 pixels to run through the model, then upscale the segmentation mask. This works, but results in less clean/lower resolution of the actual segmentation.
It looks like the HiT-SR model (available on HuggingFace) should enable higher resolution segmentation. I think at the moment, all the models downscale the image to e.g. 640x640 pixels to run through the model, then upscale the segmentation mask. This works, but results in less clean/lower resolution of the actual segmentation.