---
license: cc-by-nc-4.0
pretty_name: CS2-10k
task_categories:
  - other
tags:
  - counter-strike
  - cs2
  - gaming
  - egocentric
  - first-person
  - video
  - world-models
  - imitation-learning
  - action-prediction
  - webdataset
size_categories:
  - 100K<n<1M
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/*/*.tar
---

# CS2-10k: A Large-Scale Egocentric Counter-Strike 2 Dataset

<p align="center">
  <a href="https://huggingface.co/spaces/RekaAI/CS2-10k-viewer"><img src="https://huggingface.co/datasets/RekaAI/CS2-10k/resolve/main/viewer-banner.png" alt="Dataset Interactive Viewer"></a>
</p>

![CS2-10k preview](preview.avif)

**CS2-10k** is a large-scale egocentric gameplay dataset built from professional CS2 matches. It contains **600,000+ player-round videos** spanning **10,000+ hours** of **first-person footage**, paired with **per-frame annotations** covering **keyboard state, mouse movement, and 3D player trajectory**.

CS2-10k is built from public professional match demos sourced from [HLTV](https://www.hltv.org/). For each demo, we render first-person video at **720p, 48 fps** using the demo replay tool inside CS2, producing one video per player per round. Alongside each video, we store a `.parquet` file containing per-frame annotations synchronized to the video timeline.

The dataset is published in [WebDataset](https://github.com/webdataset/webdataset) format: `~2 GB` tar shards under `data/<map>/`, where each sample is a `<uuid>.mp4` video paired with its `<uuid>.parquet` annotations. A top-level `index.parquet` lists every clip with its `shard` location for lookup and filtering.

### Annotation schema

Every video clip has its corresponding annotations stored in a `.parquet` file:

| Field | Type | Description |
|---|---|---|
| `map` | string | Map name (e.g. "mirage", "dust2") |
| `round_number` | int | Round within the match |
| `team` | int | 0 = Terrorist, 1 = Counter-Terrorist |
| `num_frames` | int | Total frames in the clip |
| `fps` | float | Video frame rate (48.0) |
| `total_time` | float | Clip duration in seconds |
| `fov` | float | Camera field of view (90.0°) |
| `frame_data` | list[dict] | Per-frame annotation array (see below) |

### Per-frame annotations

Each entry in `frame_data` contains:

| Field | Description |
|---|---|
| `actions` | Concatenated active keys: **W/A/S/D** (movement), **J** (jump), **C** (crouch), **R** (walk), **V** (freefall), **[** (fire), **]** (scope/secondary), **-** (no input) |
| `mouse_x_delta` | Horizontal camera delta — proxy for mouse X movement |
| `mouse_y_delta` | Vertical camera delta — proxy for mouse Y movement |
| `position_x / y / z` | Player world position in game units |
| `rotation_yaw` | Camera yaw angle (−180° to 180°) |
| `rotation_pitch` | Camera pitch angle (−90° to 90°) |

## License

Released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) (attribution, non-commercial). The underlying match demos are the property of their respective rights holders; this dataset is provided for research purposes.

## Citation

```bibtex
@misc{cs2-10k,
  title  = {CS2-10k: A Large-Scale Egocentric Counter-Strike 2 Dataset},
  author = {Reka AI},
  year   = {2026},
  url    = {https://huggingface.co/datasets/RekaAI/CS2-10k}
}
```