Clarification on use of ABO metadata (item dimensions) in experiments

[lltinghuang]

Hi, thanks for releasing this interesting work!

I have a question regarding the code in scene/__init__.py. Since the work follows NeRF2Physics and uses the ABO dataset, the check os.path.exists(os.path.join(args.source_path, "transforms.json")) will always succeed. However, due to a very minor format parsing error in sceneLoadTypeCallbacks inside the try block, an exception will always be triggered. This behavior confused me, as I don’t think this is a common or meaningful logic.

Furthermore, once the exception is raised, the code calls the readABOInfo() function, which initializes points using the item_dimensions from the metadata file filtered_metadata.json (as indicated by the comment “We create random points according to abo metadata”). I was wondering if this initialization could introduce unfairness, because filtered_metadata.json is specific to your released code, while the paper neither mentions using such metadata for initialization nor explains how this file was generated.

May I kindly ask whether the experimental results in the paper also relied on item dimensions from metadata, as implemented in the released code? Please correct me if I have misunderstood, and I apologize in advance if my question is based on a misinterpretation.