Skip to content

Question about preparing real-geometry training data #2

Description

@wnn2000

Hi, thanks for the great work.

I have a question about preparing real-geometry training data. From the code, ScanNet/ScanNet++/ARKitScenes reconstruction samples seem to require preprocessed scene directories with RGB frames, depth, poses/intrinsics, and metadata.

However, VSI-590K seems to provide only RGB videos for these datasets, e.g. {"video": "scannet/scene0191_00.mp4"}, which are loaded as normal VQA videos.

Should users separately build these real-geometry scene directories and manually add corresponding sample like{"video": "processed_scannet_f/scans_train/scene0191_00", "loading_type": "rec"} entries to the training JSONL? Is there an expected preprocessing script or data format for this?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions