It seems the segmentation texture only works on orthographic camera, but is there a way to segment to use it on perspective camera, to be able to - say - occlude a world 3D scene using foot segmentation?
Yes, it is possible to use the segmentation texture with a perspective camera. You can use the segmentation texture to mask out certain areas of the 3D scene, similar to how you would use it with an orthographic camera.
To do this, you will need to apply the segmentation texture as a mask to the materials of the objects in your 3D scene. You can do this by creating a new material and assigning the segmentation texture to the "Opacity Mask" parameter in the material's properties.
Then, you can apply this material to the objects in your 3D scene that you want to be occluded based on the segmentation data. The objects will be rendered as normal, but the areas of the objects that are masked out by the segmentation texture will not be visible.