The Use of Single Camera Option

[reagan99]

It seems that the --ImageReader.single_camera 1 option might have been used during COLMAP processing in convert.py. This option forces all cameras to share identical intrinsic parameters, including the focal length, which removes variability across cameras. While this might be a default setting within convert.py (as it is part of the standard 3DGS implementation), it could potentially differ from the approach described in the paper, which emphasizes the importance of considering diverse focal lengths and distances for Gaussian size and sampling interval calculations (e.g., ( v = max(f/d))).

If this option was applied, it might have some limitations, such as reducing the accuracy of Gaussian adjustments by ignoring differences in intrinsic camera parameters. However, there’s also a possibility that the single camera option was simply left in the script by default and does not reflect the actual COLMAP workflow you used.

---

If the single camera option was used, it might:

• Limit the accuracy of Gaussian adjustments, especially when dealing with data from multiple cameras with varying focal lengths.
• Introduce a mismatch with the methodology described in the paper, which accounts for variability in focal lengths to optimize Gaussian size and sampling intervals.

On the other hand, if this option was not actively used during your COLMAP processing, its presence in the script might cause some confusion for others who review or reuse the code, as it could suggest a simplified camera setup.

---

Would it be possible to check if the single camera option was used during your COLMAP processing? If this option was applied but not necessary, it might be worth considering whether to adjust or remove it to better reflect the diversity of camera parameters as emphasized in the paper. Alternatively, if this setting was not part of your actual workflow(do not use convert.py), documenting this could help clarify its role in the script and ensure alignment with the methodology for anyone referencing or extending this work.

Thank you for considering.