about generate input ground map

[ahtso]

Hi Author,

Thanks for your great work,
there is a question about generating the input ground map.
I follow here
and try to generate ground map of KITTI.
But the output is not as my expected.

I remove the flip, scales and rotation part, so the function of get_ground becomes more simple.

    def get_ground(self, folder, flip, transform, out_intrinsic):
        c = self.calibrations[folder]
        cam_to_world = c["cam_to_world"]

        ### if transform is None:
        transform = np.eye(4)
        transform[:3, :3] = c["intrinsic"] @ np.linalg.inv(c["intrinsic_scaled"]) ### it makes transform matrix becomes unit matrix

        w, h = self.width, self.height

        cam_to_world = cam_to_world @ transform ### the extrinsic matrix would not change.
        cam_to_world[2, 3] += 1.65
        u, v = np.meshgrid(range(w), range(h), indexing="xy")
        ground = -cam_to_world[2, 3] / (cam_to_world[2, 0] * u + cam_to_world[2, 1] * v + cam_to_world[2, 2])

        ground[ground < 0] = 0
        ground = ground[None]
        ground = torch.from_numpy(ground.astype(np.float32))

        return ground, cam_to_world[2, 3], torch.tensor(cam_to_world).float()

All I got the ground map becomes all black or nothing.
Can you help to figure out?
Thank you!