Unexpectedl performance on ShapeNet55 partial inputs during inference

[0tae-87]

Hi there,

First of all, thank you for this incredible research and for sharing the code.

I am currently testing your network, but I've encountered an issue where the performance on ShapeNet55 partial inputs is lower than expected.

Here are the details of my environment:

Script used: tools/inference.py (provided in the repository)

Weights: Pre-trained c55 weights (ShapeNet55-34)

Dataset: Prepared exactly as described in the DATASET.md (unzipped and renamed folders accordingly)

Despite following the instructions, the reconstruction results seem a bit off/unstable. Could you provide some insight into what might be causing this? For instance, are there any specific preprocessing steps or scale-related issues I should double-check?

Attached files are mug and chair.
But, additionally airplane was fine.
Is it the problem caused by the number of datasets?

Thank you for your help!