Question regarding CoMA data preprocessing (low accuracy) and Bosphorus reconstruction

[changjin-cpu]

Hi @batuceng,

Thank you for open-sourcing your code for the paper "ε-Mesh Attack". It is excellent work.

I am currently working on reproducing your results. I have successfully reproduced the results on the FaceWarehouse dataset using your provided code. However, I am encountering difficulties with the CoMA and Bosphorus datasets.

1. Issue with CoMA Dataset

I downloaded the raw dataset from the official CoMA website. After converting the data to .stl format and running the evaluation, I am getting a clean accuracy of only 12.5% (0.125), which is significantly lower than the results reported in the paper.

I suspect there might be a mismatch in my preprocessing pipeline.

• Could you specificy how you converted the raw CoMA meshes into the .stl files used in your dataloader?
• Are there specific frame indices (besides the peak + 4 adjacent frames mentioned in the paper) or axis alignments required during the conversion?

2. Bosphorus Dataset

Regarding the Bosphorus dataset, the paper mentions using Poisson surface reconstruction to estimate the meshes.

• Could you share the specific script or tool (e.g., MeshLab parameters or Open3D method) used for this reconstruction? I am observing slight deviations in my results and want to ensure my mesh estimation is consistent with yours.

It would be incredibly helpful if you could share the preprocessing scripts for these two datasets.

Thanks in advance for your help!``

[changjin-cpu]

Can you share the preprocessed versions of the dataset for reproduciblity?