What coordinate system is the generated mesh based on?

[uguirgl-39]

I extracted meshes from images using SAM 3D Body.

I'm generating OBJ files based on the code from https://github.com/facebookresearch/sam-3d-body/blob/main/notebook/demo_human.ipynb:

if outputs:
    # Get image name without extension
    image_name = os.path.splitext(os.path.basename(image_path))[0]

    # Create output directory
    output_dir = f"output/{image_name}"

    # Save all results (PLY meshes, overlay images, bbox images)
    ply_files = save_mesh_results(img_cv2, outputs, estimator.faces, output_dir, image_name)

    print(f"\n=== Saved Results for {image_name} ===")
    print(f"Output directory: {output_dir}")
    print(f"Number of PLY files created: {len(ply_files)}")

else:
    print("No results to save - no people detected")

My question: I'm not sure about the orientation and coordinate system of the generated mesh data.

Specifically, I'd like to understand:

・What coordinate system is used (camera coordinates, world coordinates, etc.)?
・Which direction is the mesh facing (which axis represents "forward")?
・What is the reference point/origin?
Am I missing some documentation about this? Any clarification would be greatly appreciated!

[uguirgl-39]

And I assume that the coordination is based on the camera angle.
Am I correct???

[lucanunz]

I had similar doubts here 65. Now I have open sourced my code where, if you have the extrinsics and intrinsics of the matrix that took the rgb-d picture, then you can have the mesh/pointcloud in 3D in the frame of your choice.

Hope it can be helpful!
https://github.com/lucanunz/sam-3d-body-calibrated