How could I segment object without need to resize the whole image for it

[VictorGimenez]

I am following the example from the official webpage to use Segment Anything tool: https://keras.io/guides/keras_cv/segment_anything_in_keras_cv/ but I am using a different image from the example (1200, 1800, 3) where my image has (1920, 1080, 3) and there are another ones with percentual reduction from this one of (1920, 1080, 3), I am trying to maintain the same width and height from the input image to have the real contours/segmentation coordinates (from the original image) and to establish the input_point and input_label on the original resolution (width, height) without the need to resize through the inference_resizing method.

I have tried to replace 1024 by another width values such as 1920, 1728, 1536... or suppose that I have a "x" value it would be this "x" value instead this 1024 but I always get error on the outputs dictionary where I set the "images", "points" and "labels" dicts.

What should I do to maintain the same width, height to segment my image to have the real x and y coordinates used to segment my object?