README.md

---
title: Dataset Viewer
emoji: 🎬
colorFrom: blue
colorTo: gray
sdk: gradio
sdk_version: 6.19.0
app_file: app.py
pinned: false
---

# Dataset Viewer

Visualise episodes from:
- **EgoVerse** (`angkul07/egoverse-pick-tasks`) — zarr v3, front camera
- **ABC Teleop** (`angkul07/abc-ego`) — mcap, three cameras side by side

## Usage

1. Select a dataset from the navbar dropdown
2. (ABC Teleop only) Select a task
3. Pick a UUID from the dropdown or type one in
4. Click **Convert →** — videos stream in as each camera is ready

## Deployment

Push `app.py`, `requirements.txt`, and this `README.md` to a HuggingFace Space
(SDK: Gradio, hardware: CPU Basic is sufficient).

The Space needs read access to both dataset repos. If repos are private,
add a `HF_TOKEN` secret in the Space settings and set:

```python
os.environ["HF_TOKEN"] = os.getenv("HF_TOKEN", "")
```

at the top of `app.py`, then pass `token=os.environ["HF_TOKEN"]`
to `hf_hub_download` and `HfFileSystem()`.

## Dependencies

ffmpeg must be available in the Space environment. Add to `packages.txt`:

```
ffmpeg
```