README.md

---
pretty_name: VSIBench
license: unknown
task_categories:
- visual-question-answering
---

# VSIBench

This dataset upload contains the contents of:

`/scratch/dwirtz1/qic/spatialcode_data`

The original directory contains 1,841,072 files across:

- `arkitscenes/`
- `scannet/`
- `scannetpp/`

To keep the Hugging Face dataset repository usable, the data is stored as split
`.tar.zst` archives instead of uploading the raw file tree.

## Storage Requirements

The uploaded split archives take about `119G` on disk (`126.9GB` decimal).
After extraction, the restored `spatialcode_data/` directory takes about `127G`
on disk (`130.9GB` decimal).

Plan for:

- `119G` to download and store only the compressed archive parts.
- `127G` for the fully extracted `spatialcode_data/` directory.
- About `246G` total if you keep both the downloaded archives and extracted
  data at the same time.
- `270G+` free space is recommended to leave room for filesystem overhead and
  partial downloads.

If disk space is tight, restore one source at a time and delete its archive
parts after verifying extraction.

## Download

Install the Hugging Face CLI if needed:

```bash
pip install -U huggingface_hub
```

Download the dataset files into a local folder:

```bash
hf download qicq1c/VSIBench \
  --repo-type dataset \
  --local-dir VSIBench \
  --local-dir-use-symlinks False
```

This creates a folder like:

```text
VSIBench/
  README.md
  SHA256SUMS
  arkitscenes.tar.zst.part-000
  ...
  scannet.tar.zst.part-000
  ...
  scannetpp.tar.zst.part-000
  ...
```

## Verify

Check that all downloaded archive parts match the uploaded checksums:

```bash
cd VSIBench
sha256sum -c SHA256SUMS
```

## Restore Directory Structure

The archives contain the original top-level directories. To recreate the
original `spatialcode_data` directory structure:

```bash
cd VSIBench
mkdir -p spatialcode_data

cat arkitscenes.tar.zst.part-* | tar --zstd -xf - -C spatialcode_data
cat scannet.tar.zst.part-* | tar --zstd -xf - -C spatialcode_data
cat scannetpp.tar.zst.part-* | tar --zstd -xf - -C spatialcode_data
```

After extraction, the restored layout is:

```text
VSIBench/
  spatialcode_data/
    arkitscenes/
    scannet/
    scannetpp/
```

This matches the contents of the original source directory:

```text
/scratch/dwirtz1/qic/spatialcode_data/
  arkitscenes/
  scannet/
  scannetpp/
```

If you want to restore directly to a specific absolute path, create that
directory and use it as the extraction target:

```bash
mkdir -p /scratch/dwirtz1/qic/spatialcode_data
cat arkitscenes.tar.zst.part-* | tar --zstd -xf - -C /scratch/dwirtz1/qic/spatialcode_data
cat scannet.tar.zst.part-* | tar --zstd -xf - -C /scratch/dwirtz1/qic/spatialcode_data
cat scannetpp.tar.zst.part-* | tar --zstd -xf - -C /scratch/dwirtz1/qic/spatialcode_data
```

## Restore One Subset

You can also restore only one source:

```bash
mkdir -p spatialcode_data
cat scannet.tar.zst.part-* | tar --zstd -xf - -C spatialcode_data
```