---
title: Opencs2
canonical_url: "https://www.modelscope.cn/datasets/meisah111/Opencs2"
md_url: "https://www.modelscope.cn/datasets/meisah111/Opencs2.md"
repository: meisah111/Opencs2
chinese_name: "CS2视频数据集"
last_updated: 2026-06-27
license: "Apache License 2.0"
downloads: 260
stars: 0
---

# Opencs2

> Opencs2 - meisah111 在 ModelScope 开源的数据集。This is an unofficial ModelScope mirror of Julien Blanchon’s OpenCS2 dataset. Original author: Julien Blanchon. Original source: Hugging Face blanchon/opencs2_dataset_wds. License: CC-BY-4.0 for renders and metadata.…

meisah111/Opencs2 是 ModelScope 魔搭社区上的数据集，采用 Apache License 2.0 许可。

- **Repository**: meisah111/Opencs2
- **License**: Apache License 2.0
- **Downloads**: 260
- **Stars**: 0
- **Last updated**: 2026-06-27

Source: https://www.modelscope.cn/datasets/meisah111/Opencs2

---

# OpenCS2 WDS Dataset - ModelScope Mirror

This repository is a ModelScope mirror/re-upload of the original Hugging Face
dataset `blanchon/opencs2_dataset_wds`, uploaded with the original author's
permission.

We are not the original creators of this dataset. For the authoritative
description, latest updates, usage instructions, license, and citation, please
refer to the original Hugging Face repository:

https://huggingface.co/datasets/blanchon/opencs2_dataset_wds

Original repository: https://huggingface.co/datasets/blanchon/opencs2_dataset_wds

License: CC-BY-4.0, following the original repository.

Mirrored build:

- 165,270 POV samples
- 2,574 uncompressed tar shards
- 2,974.2 POV video hours
- 528.0 synced round-timeline hours

The dataset layout is preserved from the original WebDataset release:

```text
shards/
  opencs2-<match_id>-<map_name>-<shard>.train.tar
index/
  wids_train.json
  wds_samples.parquet
  wds_shards.parquet
  matches.parquet
  rounds.parquet
  pov_rounds.parquet
events/
  kills.parquet
  duels.parquet
  clip_events.parquet
  round_player.parquet
metadata/
  enums.parquet
ticks/
  match_id=<id>/map_name=<map>/ticks.parquet
```

The tar shards are plain `.tar` files, not `.tar.gz`. Do not repackage them:
the parquet indexes contain byte offsets into these tar files for sparse random
access.