--f_mm argument in demo.py rejects float values (type=int)

[kleer001]

First off — great work on GVHMR! The world-grounded recovery is a real step forward for monocular mocap pipelines.

I ran into a small issue when passing focal lengths from a COLMAP camera solve to demo.py. The --f_mm argument is defined as type=int in argparse, which rejects float values:

demo.py: error: argument --f_mm: invalid int value: '51.89029508963282'

Camera solves (COLMAP, RealityCapture, etc.) naturally produce float focal lengths, and even real-world lenses aren't always round numbers (e.g. 50.6mm vintage primes, or any anamorphic equivalent).

Where

tools/demo/demo.py, argument parser:

parser.add_argument("--f_mm", type=int, default=None, ...)

Suggested Fix

Change type=int to type=float:

parser.add_argument("--f_mm", type=float, default=None, ...)

The downstream usage in create_camera_sensor() is a float multiplication (diag_img / diag_fullframe * f_fullframe), so this should be a safe change with no side effects. The default fallback path (np.random.choice([24, 26, 28, ...])) also works fine as float.

Workaround

For anyone hitting this in the meantime — rounding '51.89029508963282' to int (--f_mm 52) works but loses precision from your camera solve.

Thanks again for the excellent project!