---
license: other
---

# Internal export: RGB + ground-truth intrinsics (validation split)

**Private use.** This repo is for sharing data via Hugging Face. Keep the HF repo **private** and only grant access to colleagues who may use the underlying data.

## What this is

RGB JPEGs from the per-scene **subset_106** sharp-frame folder (up to ~106 images per scene) plus
**ground-truth camera intrinsics** (`pointcloud_intrinsics.npz`) for the same scenes. Scene IDs
come from from the validation metadata in the [official VSI-benchmark git repository](https://github.com/vision-x-nyu/thinking-in-space/tree/main/data/meta_info) :

- `scannet_meta_info_val.json`
- `arkitscenes_meta_info_val.json`
- `scannetpp_meta_info_val.json`


## Layout

- **Images:** `<dataset>/<scene_id>/subset_106/*.jpg`  
  (`dataset` ∈ `scannet`, `arkitscenes`, `scannetpp`)
- **Intrinsics:** `gt_intrinsics/<dataset>/<scene_id>/pointcloud_intrinsics.npz`  

Intrinsics rows follow the **same order** as sorted RGB filenames in `subset_106`. Each timestep
is a 3×3 `K`; arrays use the Pi3-style leading batch dimension inside the `.npz`.

## Legal

You must comply with **ScanNet**, **ARKitScenes**, and **ScanNet++** license terms. This export is
a convenience copy for collaborators; it is not affiliated with or endorsed by the official
VSI-Bench release.


---