CS2-10k

CS2-10k: A Large-Scale Egocentric Counter-Strike 2 Dataset 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. 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 format: `~2 GB` tar shards under `data/ /`, where each sample is a ` .mp4` video paired with its ` .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