Question about evaluation on MP3D: Pixel location is outside the image.

[Anonymous-N]

Thank you for your outstanding work and for open-sourcing the code. I have successfully reproduced the results from the paper on the HM3D dataset. However, when evaluating on the MP3D dataset, I encountered the following error.

File "/home/user/Workspace/vlfm/vlfm/utils/img_utils.py", line 43, in place_img_in_img
    assert 0 <= row < img1.shape[0] and 0 <= col < img1.shape[1], "Pixel location is outside the image."
AssertionError: Pixel location is outside the image.

BrokenPipeError: [Errno 32] Broken pipe
 70%|████████████████████████████▋            | 1533/2195

Reading the code, specifically the part where the map is updated and overlaid, is there a need to add specific boundary restrictions or constraints for the MP3D dataset? I wonder if the related code has not been fully disclosed.

        # Determine where this mask should be overlaid
        cam_x, cam_y = tf_camera_to_episodic[:2, 3] / tf_camera_to_episodic[3, 3]

        # Convert to pixel units
        px = int(cam_x * self.pixels_per_meter) + self._episode_pixel_origin[0]
        py = int(-cam_y * self.pixels_per_meter) + self._episode_pixel_origin[1]

        # Overlay the new data onto the map
        curr_map = np.zeros_like(self._map)
        curr_map = place_img_in_img(curr_map, curr_data, px, py)

Or is it possible that it's an issue with the configuration file? For example, the predefined value map size of 1000 might only be suitable for HM3D? In the provided files, I only saw configurations related to HM3D, but none specific to MP3D.

I would appreciate any advice you could provide. Thank you.