Refresh the official English dataset card

README.md

@@ -1,7 +1,7 @@
 ---
 language:
 - en
-pretty_name: CoCode World Data
+pretty_name: CoCode-World Data
 size_categories:
 - 1K<n<10K
 tags:
@@ -18,28 +18,32 @@ configs:
     path: train/train-*.tar
 ---
 
-# CoCode World Data
+# CoCode-World Data
 
-CoCode World Data provides **5,000 aligned training clips**, each with a GTA target video, a proxy555 control video, the original **caption method v1** text, and a frame-level alignment record. The release also includes the recorded map geometry, player controls, camera records, world state, and proxy compiler source needed to inspect or render the captured scenes in other proxy styles.
+**Synchronized multi-view video data for multi-player world models**
 
-本数据集包含 **5000 组训练 clip**：proxy555 视频、target 视频及原始 caption v1，并提供地图建模、玩家操作、相机、world state 和帧级对齐元数据，方便社区设计自己的 proxy 渲染风格。保留原片：每段 124 帧、24 FPS，实际时长约 **5.17 秒**。
+[Project Page](https://iapple233.github.io/CCW/) · [Code](https://github.com/IApple233/CoCode-World) · [Model](https://huggingface.co/iapple1/CoCode-World-Model)
 
-## Training data
+## Dataset overview
 
-| Property | Value |
-|---|---|
-| Split | train; exactly 5,000 distinct clip IDs |
-| Video layout | Horizontal dual view: agent_000 left, agent_001 right |
-| Resolution | 2560 × 720 pixels; 1280 × 720 per view |
-| Frames and playback | 124 frames at 24 FPS; 5.166667 seconds |
-| Proxy compiler | 5.5.1, also referred to as proxy555 |
-| Captions | Original caption method v1, in English |
-| Packaging | 25 uncompressed TAR shards, 200 examples per shard |
-| Recording coverage | 21 episodes |
+CoCode-World Data contains **5,000 paired training examples** for synchronized two-player video generation. Each example includes a target RGB video, a corresponding 3D proxy video, an English caption, and frame alignment metadata.
+
+The release also provides scene geometry, player actions, camera parameters, world-state observations, and a proxy rendering toolkit for exploring alternative scene representations.
 
-The common “5-second clip” description refers to these original 124-frame clips. Videos and caption text retain their source bytes; they are not trimmed or regenerated. Each video contains both cameras. Thus 5,000 examples contain 5,000 proxy videos and 5,000 target videos, each with two views.
+## Dataset structure
+
+| Property | Description |
+|---|---|
+| Split | Training |
+| Examples | 5,000 |
+| Camera layout | Two views arranged side by side |
+| View order | `agent_000` on the left; `agent_001` on the right |
+| Resolution | 2560 × 720; 1280 × 720 per view |
+| Duration | 124 frames at 24 FPS, approximately 5.17 seconds |
+| Captions | English scene descriptions and per-view timelines |
+| Storage | 25 TAR shards, each containing 200 examples |
 
-Each sample uses the same stable ID across its archive members:
+Each example contains the following files:
 
 ```text
 <sample_id>.proxy.mp4
@@ -49,44 +53,41 @@ Each sample uses the same stable ID across its archive members:
 <sample_id>.json
 ```
 
-`manifests/train.jsonl` provides the complete sample index, archive/member references, captions, source identities, and checksums. `manifests/train.parquet` provides the same 5,000 records in columnar form. Selection and validation evidence are distributed beside the manifest. The example reader in `examples/` works directly with the archives and does not require a video decoder merely to retrieve a sample. See [the usage guide](docs/USAGE.md) for verified standard-library extraction and Hugging Face Datasets streaming examples.
-
-## Selection and caption identity
-
-The source collection contains 8,552 paired clips. Its existing automated quality screen marks 8,318 as `keep`; this release selects exactly 5,000 from that pool with proportional episode stratification and deterministic selection. Flagged and review cases are excluded. This is an automated quality screen, not a claim of complete manual review. Source-segment identities are checked against the existing held-out evaluation segments before release.
-
-“Caption v1” denotes the original caption method, with an overall summary plus left- and right-view timelines. The public annotation schema is `caption-v1`. Caption text retains its original bytes. A public reference prompt, with release naming conventions, is documented in [the usage guide](docs/USAGE.md).
+The sample index is available as `manifests/train.jsonl` and `manifests/train.parquet`. It links each example to its shard, media, captions, and metadata.
 
-## Metadata and reconstruction
+## Download and use
 
-See [`metadata/README.md`](metadata/README.md), [`metadata/catalog.json`](metadata/catalog.json), and the guides in [`docs/`](docs/) for archive contents and reconstruction steps. The metadata covers all 21 recording episodes, including intervals beyond the selected 5,000 training clips. This preserves contiguous state and control histories for reuse.
+```bash
+pip install -U huggingface_hub
+hf download iapple1/CoCode-World-Data --repo-type dataset \
+  --include 'train/*' 'manifests/train.jsonl' \
+  --local-dir datasets/CoCode-World-Data
+```
 
-The metadata includes:
+The [usage guide](docs/USAGE.md) provides extraction and streaming examples. To prepare the data for CoCode-World training, use the official [code repository](https://github.com/IApple233/CoCode-World):
 
-- Static map placements, geometry, collision information, and material metadata.
-- Both players' recorded actions, button states, motion, camera parameters, and timing records.
-- World-state observations of agents, vehicles, pedestrians, objects, and environment changes.
-- Frame/source mappings, cross-view timing, prepared scene records, and compiler provenance.
-- The offline proxy compiler runtime and Python dependency records, with documented coordinate, timing, and rendering conventions.
+```bash
+python scripts/prepare_training_data.py \
+  --dataset-dir datasets/CoCode-World-Data \
+  --output-dir datasets/prepared
+```
 
-Use the per-clip alignment record and global frame index to join a training video to its original observations. The two cameras must be matched using the recorded server clock; identical local frame numbers do not establish a cross-camera match. Preserve left/right camera ownership when building new proxy representations.
+See the [training guide](https://github.com/IApple233/CoCode-World/blob/main/docs/TRAINING.md) for caching and model training.
 
-Alternative colors, materials, primitive geometry, and visualizations can be derived from the recorded scene and camera data. The metadata describes captured trajectories and actions. It does not itself provide a game simulator that can predict a new world-state sequence after arbitrary new player actions.
+## Scene metadata and rendering
 
-## Public identifiers and privacy
+Additional metadata covers 21 recording episodes and includes scene geometry, player controls, camera trajectories, world-state observations, and frame mappings. The accompanying proxy rendering toolkit supports alternative colors, materials, and geometric representations.
 
-Training examples use `clip_000001` through `clip_005000`; recordings use `episode_001` through `episode_021`. Map assets are published as `scene_001`. These public identifiers consistently link video, caption, controls, scene state, and frame indices.
+See the [metadata overview](metadata/README.md) and [documentation](docs/) for the available files and reconstruction workflow. Match the two views using the recorded timestamps and alignment records, and preserve their left/right camera assignment.
 
-Public metadata normalizes internal paths and collection identifiers, removes local account and machine details, and omits recording wall-clock identity fields. Monotonic timestamps, frame identities, in-game time, camera parameters, player actions, and geometric state needed for synchronization and rendering are preserved. Original video and caption bytes are retained. The private identifier mapping is not distributed.
+## Collection and selection
 
-## Verification
+The release is a selected subset of a larger paired-video collection. Examples were selected after automated quality screening, with coverage balanced across recording episodes. The target videos, proxy videos, and English captions are provided as recorded. The release is intended for training; the separate [evaluation examples](https://github.com/IApple233/CoCode-World/tree/main/examples/evaluation) are provided in the code repository.
 
-The release includes sample checksums, archive checksums, counts, selection evidence, and validation reports. To verify the published files after download, run:
+## Intended use and limitations
 
-```bash
-sha256sum -c SHA256SUMS.txt
-```
+The dataset supports research on multi-player world models, multi-view video generation, and scene-conditioned rendering. Automated screening does not guarantee that every example is free of artifacts. Scene metadata describes recorded trajectories and actions; it does not provide a simulator for arbitrary new action sequences.
 
-Checksums apply to the exact released bytes. Video and caption hashes verify the preserved source files. Alignment and sample-metadata hashes verify the public records after path and identifier normalization.
+## Data rights
 
-Release prepared on 2026-11-11.
+Game-derived videos and assets remain subject to the rights and terms of their respective owners. Software included with the dataset retains its accompanying license. Users are responsible for ensuring that their intended use complies with the applicable terms.

SHA256SUMS.txt

@@ -1,5 +1,5 @@
 1e33470154bfa09a7bc325159d169e6886377c5bf8ed796d15e90212b9507746  .gitattributes
-dcd7b2c1b4c5862a42c68d5433b826a3d7e4b4526f6e0b0f5a7dbaa6db32a64a  README.md
+6c774128a457cefd9a1a4a67895cbf8fb9e8230c0f775b60ec4449d128132959  README.md
 89245dc3c589b3f05e2a4e39c0a877e9bfd2c654c914b8574621c1def9efe9b4  docs/RENDERING.md
 e7cc21745d32dc4d4a89bbf27e85ad192291bca42544bd52ebae87e5394afcf0  docs/USAGE.md
 57c1f2626720344b12d3f67a424929c574addac3425a9d1dabaa8d8edd4e0239  docs/caption_v1_prompt.txt