Add files using upload-large-folder tool

README.md

@@ -0,0 +1,213 @@
+---
+license: cc-by-nc-4.0
+task_categories:
+  - video-classification
+  - image-to-image
+  - text-to-video
+  - other
+language:
+  - en
+tags:
+  - g-buffer
+  - inverse-rendering
+  - forward-rendering
+  - intrinsic-decomposition
+  - video
+  - benchmark
+  - synthetic
+  - games
+  - neurips-2026
+pretty_name: Generative World Renderer — Reviewer Sample (40 clips)
+size_categories:
+  - n<1K
+configs:
+  - config_name: default
+    data_files:
+      - split: sample
+        path: metadata.csv
+---
+
+# Generative World Renderer — Reviewer Sample (40 clips)
+
+> **NeurIPS 2026 Evaluations & Datasets Track — anonymous submission.**
+> This repository hosts a **40-clip sample** of the dataset described in our
+> submission *"A Scaling Recipe for Generative World Renderer"*. The full
+> dataset (≈ 4 M frames, 720p / 30 FPS) exceeds the 4 GB threshold and will be
+> released in full upon acceptance; this sample is provided so reviewers can
+> inspect the data without personal requests.
+
+## Summary
+
+A continuous, multi-channel **G-buffer × RGB video dataset** captured from two
+visually rich AAA games via a non-intrusive, graphics-API-level interception
+pipeline. Each clip provides synchronized:
+
+- **RGB** (the rendered frame)
+- **depth** (linearized depth buffer)
+- **normal** (camera-space surface normals, reconstructed from depth)
+- **albedo** (base color)
+- **metallic** (PBR metallic channel)
+- **roughness** (PBR roughness channel)
+- **ambient_occlusion** *(Cyberpunk 2077 only)*
+
+All channels are encoded as MP4 (H.264, near-lossless) at 720p / 30 FPS.
+
+## Sample composition
+
+| Subset             | Clips | Per-clip channels | Source                            |
+|--------------------|------:|-------------------|-----------------------------------|
+| `cyberpunk2077`    |    20 | 7                 | Cyberpunk 2077 (CD Projekt Red)   |
+| `black_myth_wukong`|    20 | 6                 | Black Myth: Wukong (Game Science) |
+
+Clips were **randomly sampled with session diversity**: each clip in the
+sample comes from a distinct capture session, covering different scenes,
+weathers, times of day, and camera/scene motion patterns. The full
+dataset additionally contains motion-blur variants and per-clip VLM-derived
+attribute labels (scene, weather, motion, dominant texture) that are not
+included in this reviewer sample for size reasons; they will accompany the
+full release.
+
+## Repository layout
+
+```
+.
+├── README.md
+├── metadata.csv                   # per-clip manifest (game, clip_id, channels)
+├── croissant.json                 # machine-readable metadata (Croissant 1.0)
+├── cyberpunk2077/
+│   ├── clip_000/
+│   │   ├── rgb.mp4
+│   │   ├── albedo.mp4
+│   │   ├── depth.mp4
+│   │   ├── normal.mp4
+│   │   ├── metallic.mp4
+│   │   ├── roughness.mp4
+│   │   └── ambient_occlusion.mp4
+│   ├── clip_001/ ...
+│   └── clip_019/
+└── black_myth_wukong/
+    ├── clip_000/
+    │   ├── rgb.mp4
+    │   ├── albedo.mp4
+    │   ├── depth.mp4
+    │   ├── normal.mp4
+    │   ├── metallic.mp4
+    │   └── roughness.mp4
+    ├── clip_001/ ...
+    └── clip_019/
+```
+
+All channels within a clip are temporally aligned frame-by-frame.
+
+## How to load
+
+```python
+from huggingface_hub import snapshot_download
+
+local = snapshot_download(
+    repo_id="anonymous111111/generative-world-renderer-clips",
+    repo_type="dataset",
+)
+# then iterate cyberpunk2077/clip_*/ and black_myth_wukong/clip_*/
+```
+
+You can decode the per-channel videos with any standard tool (e.g.
+`torchvision.io.read_video`, `decord`, `imageio`, or `ffmpeg`).
+
+## Intended evaluative role
+
+In the spirit of the NeurIPS 2026 E&D track, we make explicit **what claims
+this contribution supports and under which assumptions**.
+
+**Supports**:
+
+1. *Cross-dataset generalization of inverse renderers* — fine-tuning on this
+   dataset improves geometry and material decomposition on real-world videos
+   compared to training only on existing synthetic corpora. Tested by
+   measuring AbsRel / RMSE / threshold-accuracy (depth), PSNR / LPIPS (albedo),
+   and angular error (normals) on a held-out Black Myth: Wukong split, plus a
+   VLM-based protocol on real captures without ground truth.
+2. *G-buffer-conditioned forward synthesis* — the long, dynamic G-buffer ↔
+   RGB pairs enable training generative renderers that follow textual style
+   prompts while respecting geometry and materials.
+3. *VLM-as-judge for material evaluation* — we propose a VLM-based pairwise
+   protocol for metallic/roughness assessment in the absence of ground truth,
+   and report its correlation with human preference.
+
+**Assumptions / limitations**:
+
+- **Synthetic-to-real gap remains.** The dataset is captured from game
+  engines; while AAA titles exhibit photoreal lighting and complex
+  geometry, residual stylization and engine-specific BRDF approximations are
+  not identical to real-world physics.
+- **Camera-space normals are reconstructed from depth** (Eq. 1 in the paper)
+  rather than read directly from the engine, because we cannot reliably
+  obtain the view matrix at API level. This may smooth fine geometric
+  detail at depth discontinuities.
+- **Material packing.** Metallic and roughness are decoupled from packed
+  G-buffer targets and rendered into spatially distinct screen regions before
+  recording, to avoid inter-channel bleeding under MP4 compression. The
+  decoupled crops are recombined at load time. Compression noise is therefore
+  intra-channel rather than cross-channel, but is not zero.
+- **Coverage.** Only two titles, two engines. Generalization claims should
+  be read as: "fine-tuning on these two titles improves over training on
+  prior-art synthetic data," not as proof of universal coverage.
+- **No combat / no NPC-interaction sequences** in Wukong (deliberate, to
+  avoid action-cinematic camera artifacts).
+
+## Croissant metadata
+
+A machine-readable Croissant 1.0 description is available at
+[`croissant.json`](./croissant.json) at the root of the repository, including
+core fields (license, citation, distribution, recordSet) and **Responsible AI
+fields** (data collection, personal/sensitive information, social impact,
+known biases, use cases, and limitations).
+
+## Responsible AI
+
+- **Personal / sensitive information.** None. Video is captured from the
+  rendered output of single-player video games. No real persons, voices, or
+  personal data are recorded. NPC characters in Cyberpunk 2077 are fictional.
+- **Offensive content.** Both titles are rated for adult audiences and may
+  contain stylized violence (Black Myth: Wukong) or mature themes
+  (Cyberpunk 2077). Capture sessions deliberately avoid combat and explicit
+  content; sampled clips here focus on environmental traversal.
+- **Bias.** Scene coverage reflects the artistic choices of two specific AAA
+  titles (a near-future cyberpunk metropolis and a Chinese-mythology dark
+  fantasy). The dataset is **not** a general-purpose, demographically
+  representative video corpus and should not be used as one.
+- **Intended use.** Research on inverse and forward rendering, intrinsic
+  decomposition, G-buffer-conditioned generation, and evaluation methodology
+  for these tasks.
+- **Out-of-scope use.** Training systems for surveillance, behavior
+  prediction, content moderation of real human imagery, or any task that
+  presumes the data depicts real-world events or persons.
+
+## License & redistribution
+
+The captured pixels are derivative of copyrighted game engines and assets
+owned by their respective publishers (CD Projekt Red for Cyberpunk 2077,
+Game Science for Black Myth: Wukong). We release **only the captured
+G-buffer / RGB recordings** under **CC BY-NC 4.0 for non-commercial research
+use**. Users must comply with each publisher's End-User License Agreement
+when re-using or redistributing. The dataset must not be used to train
+models intended for commercial deployment that reproduce game-specific
+visual identity.
+
+## Citation
+
+```
+@inproceedings{anonymous2026gwr,
+  title  = {A Scaling Recipe for Generative World Renderer},
+  author = {Anonymous},
+  booktitle = {Submitted to the NeurIPS 2026 Evaluations and Datasets Track},
+  year   = {2026},
+  note   = {Under double-blind review.}
+}
+```
+
+## Contact
+
+For the duration of the review period, please use the OpenReview discussion
+on the corresponding submission. Author identity will be disclosed in the
+camera-ready version.

black_myth_wukong/clip_003/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4a6739c9cbdf80f89beb9c9c0e789427b1b49c590422e29c0799793f3fd71fc
+size 12060728

black_myth_wukong/clip_003/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d4e7f2c85c34329c52611cb8e4c4457965622ef58f51740b6b2daa022ac5bfc
+size 1061054

black_myth_wukong/clip_003/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55703b3d2773a64fb08bdf65f6830953fe9dcef1bd00725aea0ff6e319f65779
+size 2135884

black_myth_wukong/clip_003/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6d5309ecffa30fa1675e1fd54fc236ac8081546cc6024f606d1c99a108a6060
+size 6142749

black_myth_wukong/clip_003/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f3585e344ef37d6aca4e58b20c55341fba6eeb2f309702424a86566da4da1cb
+size 7234173

black_myth_wukong/clip_007/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b2baaf178e8d13d7bb65c8a2fbbcb69ba7bc05ada283d17575b9875789e18f
+size 2091577

black_myth_wukong/clip_011/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99745004a0e0af1caa43924a59529b3c3ae8668c1c30dbbed92a1aae4196ec20
+size 6466037

black_myth_wukong/clip_011/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:969ef3458520c95a492c8fc2dbdbd9b7f7d8f5d7109af21e3d901130b39434ac
+size 2429958

black_myth_wukong/clip_011/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf1e3b3ab9f4e9adb17bf7809c8a3daa9bb3f8773bbfc799445856c8e1392b6d
+size 19646784

black_myth_wukong/clip_012/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad254a920b66fec1796eba966e32daaf78de1e9a9d2bbf24f92b795ee4a7b181
+size 609095

black_myth_wukong/clip_012/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d63757216019153a94acd2bc71970c63c8b940895418877b24449c58c602f9c
+size 4890760

black_myth_wukong/clip_012/normal.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d83f304ac21eef878520214b13d720c6ed3cdd123f257f8010e8a69234d9d0a9
+size 24994375

black_myth_wukong/clip_012/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56f4b25d8a14e86fe280d46d71bb0c1291a33141695fc46715beb5d9dc543755
+size 15236168

black_myth_wukong/clip_012/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:118461a7303feb7860503afa276c1bbfd0098d2ed05cf09eae523be130d52caf
+size 24693309

black_myth_wukong/clip_013/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bb9e89e1963cf8fc2393f692ccb67749431cf1881255003e3d9984bcb42a7c6
+size 2219125

black_myth_wukong/clip_013/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10b13d711172a6ad3af697eb56f5693c45b937569047f71354f328fe4d2199dc
+size 2390597

black_myth_wukong/clip_014/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72fe4164d38902df2bded78b9160588ddeb6cc9818a7c90145284af3aa61ca40
+size 20648488

black_myth_wukong/clip_014/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a40e322c584ad5ad853463f2bb2ed693ff09b3098a24dfa1e938b2da65bbc801
+size 4224218

black_myth_wukong/clip_014/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d5d18f98ee5b1f3e0283bb0096253a041140441230e9b5d86f59f12e4f01dec
+size 3320470

black_myth_wukong/clip_014/normal.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23c76c1259546df7783b1e7da5872e99e8d50d0899e7a17956fc0ed8ac998573
+size 49496289

black_myth_wukong/clip_014/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecd5b0b0cc05989f1d425bfe34f743d4e67f1ac7fd409235a23ca7317700b4d2
+size 19539774

black_myth_wukong/clip_014/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1284e0b443d17690ece27dbc79d43fee08965e30bef0e27c5964c3d13256cafd
+size 16949768

croissant.json

@@ -0,0 +1,320 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "wd": "https://www.wikidata.org/wiki/"
+  },
+  "@type": "sc:Dataset",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "name": "generative-world-renderer-clips",
+  "description": "A 40-clip reviewer sample of a large-scale G-buffer x RGB video dataset captured from two AAA games (Cyberpunk 2077 and Black Myth: Wukong) via graphics-API-level interception. Each clip provides temporally synchronized MP4 videos for RGB and five (or six) G-buffer channels at 720p / 30 FPS. Released anonymously for double-blind review at the NeurIPS 2026 Evaluations & Datasets Track.",
+  "url": "https://huggingface.co/datasets/anonymous111111/generative-world-renderer-clips",
+  "version": "1.0.0",
+  "datePublished": "2026-05-06",
+  "license": "https://creativecommons.org/licenses/by-nc/4.0/",
+  "creator": {
+    "@type": "Organization",
+    "name": "Anonymous (NeurIPS 2026 E&D Track submission)"
+  },
+  "publisher": {
+    "@type": "Organization",
+    "name": "Anonymous (NeurIPS 2026 E&D Track submission)"
+  },
+  "keywords": [
+    "g-buffer",
+    "inverse rendering",
+    "forward rendering",
+    "intrinsic decomposition",
+    "video",
+    "synthetic dataset",
+    "AAA games",
+    "evaluation"
+  ],
+  "citeAs": "@inproceedings{anonymous2026gwr, title={A Scaling Recipe for Generative World Renderer}, author={Anonymous}, booktitle={Submitted to the NeurIPS 2026 Evaluations and Datasets Track}, year={2026}}",
+  "isLiveDataset": false,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "readme",
+      "name": "README.md",
+      "description": "Human-readable dataset card.",
+      "contentUrl": "README.md",
+      "encodingFormat": "text/markdown",
+      "sha256": "149beaa2a298ac56e735f5dc813d6f007383966c9a4338a913978d1e742f50ff"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "metadata-csv",
+      "name": "metadata.csv",
+      "description": "Per-clip manifest (game, clip_id, source session, channels).",
+      "contentUrl": "metadata.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "694d2f52df780f8700dff44d9ff78a857dcb13fb853b053fc08d90e4986b718f"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-rgb",
+      "name": "cyberpunk2077_rgb_videos",
+      "description": "Rgb channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/rgb.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-albedo",
+      "name": "cyberpunk2077_albedo_videos",
+      "description": "Albedo channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/albedo.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-depth",
+      "name": "cyberpunk2077_depth_videos",
+      "description": "Depth channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/depth.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-normal",
+      "name": "cyberpunk2077_normal_videos",
+      "description": "Normal channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/normal.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-metallic",
+      "name": "cyberpunk2077_metallic_videos",
+      "description": "Metallic channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/metallic.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-roughness",
+      "name": "cyberpunk2077_roughness_videos",
+      "description": "Roughness channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/roughness.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-cyberpunk2077-ambient_occlusion",
+      "name": "cyberpunk2077_ambient_occlusion_videos",
+      "description": "Ambient Occlusion channel MP4 videos for the Cyberpunk 2077 subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "cyberpunk2077/clip_*/ambient_occlusion.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-black_myth_wukong-rgb",
+      "name": "black_myth_wukong_rgb_videos",
+      "description": "Rgb channel MP4 videos for the Black Myth: Wukong subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "black_myth_wukong/clip_*/rgb.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-black_myth_wukong-albedo",
+      "name": "black_myth_wukong_albedo_videos",
+      "description": "Albedo channel MP4 videos for the Black Myth: Wukong subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "black_myth_wukong/clip_*/albedo.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-black_myth_wukong-depth",
+      "name": "black_myth_wukong_depth_videos",
+      "description": "Depth channel MP4 videos for the Black Myth: Wukong subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "black_myth_wukong/clip_*/depth.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-black_myth_wukong-normal",
+      "name": "black_myth_wukong_normal_videos",
+      "description": "Normal channel MP4 videos for the Black Myth: Wukong subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "black_myth_wukong/clip_*/normal.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-black_myth_wukong-metallic",
+      "name": "black_myth_wukong_metallic_videos",
+      "description": "Metallic channel MP4 videos for the Black Myth: Wukong subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "black_myth_wukong/clip_*/metallic.mp4"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "fileset-black_myth_wukong-roughness",
+      "name": "black_myth_wukong_roughness_videos",
+      "description": "Roughness channel MP4 videos for the Black Myth: Wukong subset (one file per clip).",
+      "encodingFormat": "video/mp4",
+      "includes": "black_myth_wukong/clip_*/roughness.mp4"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "clips",
+      "name": "clips",
+      "description": "One record per clip. Each clip provides several temporally synchronized MP4 videos at 720p / 30 FPS (one per G-buffer channel), enumerated as FileSets in the distribution.",
+      "key": {
+        "@id": "clips/clip_id"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "clips/game",
+          "name": "game",
+          "description": "Source game subset: cyberpunk2077 or black_myth_wukong.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "game"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "clips/clip_id",
+          "name": "clip_id",
+          "description": "Clip identifier within the subset (clip_000 .. clip_019).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "clip_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "clips/source_session",
+          "name": "source_session",
+          "description": "Capture-session timestamp from the original (full) dataset.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "source_session"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "clips/source_clip",
+          "name": "source_clip",
+          "description": "Clip identifier within its capture session in the original (full) dataset.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "source_clip"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "clips/channels",
+          "name": "channels",
+          "description": "Pipe-separated list of channel names available for this clip.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "channels"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataCollection": "Captured from the rendered output of two single-player video games (Cyberpunk 2077 and Black Myth: Wukong) using a non-intrusive ReShade-based pipeline that intercepts the graphics API at runtime, identifies the desired render-target attachments via offline RenderDoc inspection, and shades them to screen for hardware-accelerated capture via OBS at near-lossless bitrate. A dual-monitor 'mosaic' compositing strategy is used to record all six channels simultaneously at 720p, ensuring exact temporal synchronization. No game binaries, asset files, or memory contents are extracted.",
+  "rai:dataCollectionType": "Screen capture of in-engine rendered output via graphics API interception.",
+  "rai:dataCollectionRawData": "Raw recordings are MP4-encoded screen captures of the composed G-buffer mosaic. No game asset files, save files, or memory dumps are stored.",
+  "rai:dataCollectionTimeframe": "2025-11 to 2026-01.",
+  "rai:dataAnnotationProtocol": "For the full dataset, per-clip categorical attributes (scene type, weather, camera/scene motion class, dominant texture) are produced by Qwen3-VL-235B-A22B-Instruct over five uniformly sampled, timestamp-burned-in frames per clip, then audited by the authors. The full annotations are not included in this 40-clip reviewer sample.",
+  "rai:dataAnnotationPlatform": "vLLM-served Qwen3-VL-235B-A22B-Instruct.",
+  "rai:dataAnnotationAnalysis": "Annotation reliability is assessed via author spot-checks over a randomly drawn subset; the authors did not measure inter-annotator agreement because annotation is single-source (a VLM).",
+  "rai:dataPreprocessingProtocol": [
+    "Buffer identification via offline RenderDoc frame analysis followed by per-frame signature matching at runtime (format, extent, draw-call span).",
+    "Camera-space normals reconstructed from depth via inverse projection and finite differences.",
+    "Metallic and roughness decoupled into spatially distinct screen regions before recording to avoid inter-channel bleeding under MP4 compression; recombined at load time.",
+    "Center-cropping of source buffers prior to mosaic tiling to preserve aspect ratio.",
+    "Filtering of clips where both scene and camera are static, and frames with extremely low luminance.",
+    "Synthesis of motion-blur RGB variants by 8x RIFE interpolation and linear-domain temporal averaging (not included in the reviewer sample)."
+  ],
+  "rai:dataReleaseMaintenancePlan": "The reviewer sample (this repository, 40 clips, ~2.9 GB) will remain available on Hugging Face for the full NeurIPS 2026 review and post-publication period. Upon paper acceptance, the full ~4M-frame dataset will be released at the same anonymous repository (de-anonymized) under the same license. Any updates will be versioned via Hugging Face revisions.",
+  "rai:dataUseCases": [
+    "Training and evaluating video inverse renderers (depth, normal, albedo, metallic, roughness estimation).",
+    "Training G-buffer-conditioned generative forward renderers, including controllable relighting and style transfer.",
+    "Studying domain transfer from synthetic AAA-game data to in-the-wild real videos.",
+    "Benchmarking VLM-as-a-judge protocols for material-property evaluation in the absence of ground truth."
+  ],
+  "rai:dataLimitations": [
+    "Coverage is limited to two game titles and two engines; generalization claims are bounded accordingly.",
+    "Camera-space normals are reconstructed from depth rather than read directly from the engine; fine geometric detail near depth discontinuities may be smoothed.",
+    "Material channels are subject to MP4 compression noise within each channel.",
+    "Intentional exclusion of combat sequences in Black Myth: Wukong; the dataset under-represents fast camera shake.",
+    "Game stylization (artistic post-processing, engine-specific BRDF approximations) is not identical to real-world physics."
+  ],
+  "rai:dataBiases": "Scene coverage reflects the artistic choices of the two source titles: a near-future cyberpunk metropolis (urban, metal-heavy, neon-lit) and a Chinese-mythology dark-fantasy world (natural and architectural, often diffuse and shadowed). The dataset is not demographically or geographically representative of real-world environments and should not be used as such.",
+  "rai:personalSensitiveInformation": "None. All depicted characters are fictional in-game entities; no real persons, voices, faces, biometric information, or other personal data are recorded. No user-identifying telemetry is captured during the recording sessions.",
+  "rai:dataSocialImpact": "Positive: enables more rigorous evaluation of inverse and forward rendering algorithms by providing temporally continuous, multi-channel ground-truth supervision that was previously scarce. Negative risks: the high realism of AAA-game footage could be misused to train generative systems that fabricate convincing scene imagery; the CC BY-NC license and explicit non-commercial / non-deceptive use clause are intended to mitigate this. Capture is non-intrusive and does not redistribute game assets, but downstream redistribution must comply with each publisher's End-User License Agreement."
+}

cyberpunk2077/clip_001/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:388c4252e4323353c232b909d442001864e8306dfc77da567ad6fe78ce9116de
+size 9599972

cyberpunk2077/clip_001/ambient_occlusion.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9e45679f682443194e99235c08de5a674aaf1b05bf344708380a41419a93380
+size 5081907

cyberpunk2077/clip_001/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6461c32797e32e769f53265041f0d2fa1acf3091e453a4e4bd9257d963b23d2
+size 2248461

cyberpunk2077/clip_001/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66415f33656b2647d10e1676972ed7b67200fa50857683b12021f5ebd719edab
+size 4938844

cyberpunk2077/clip_001/normal.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:831b5fb4feb676911d4d60fe6bcfe16a38116a89c7affefe2df77e2d024a2aec
+size 12815479

cyberpunk2077/clip_001/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e24fa455683bc4c0272e09287130f8f6e45feb417afd5f3f353181217f6cbe00
+size 6990713

cyberpunk2077/clip_001/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3e04655b76cf2abe97702f9d2428f6b0655e123e2c8d7e9bb32aa93d31d00bd
+size 20692435

cyberpunk2077/clip_005/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:371d47ff375034d3408ade570f389fde06718488f620425c1f3b89becf6ce59d
+size 9874724

cyberpunk2077/clip_005/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc2540c17e3eebce2f493d0435e151ae29b5398c9577430e2d4a5c02ec33bd8b
+size 4496101

cyberpunk2077/clip_005/normal.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:899b2e29f98fdd6eb17882c5de2dd87b5ed2afa36fde743b9c6719c8200ed5c9
+size 11985620

cyberpunk2077/clip_005/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dedd390f7869090dd7a3c9609fa2adaefe4462cbb974685c9c234c93fed8547
+size 11126010

cyberpunk2077/clip_005/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd77d8e22b5cd85c553e1e25feb4fe0b7ca816761922f3c0d3a8952cb0e53267
+size 14188522

cyberpunk2077/clip_006/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38ae819825235da7fd3748c632855957094bbfcf0de305e0223a2cfd45acd14d
+size 10216681

cyberpunk2077/clip_006/ambient_occlusion.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abb76a729732c523c227cdab36a38b1c1a1741ac8466edd14aac951ce7fce0af
+size 3866441

cyberpunk2077/clip_006/depth.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d61bdb81ba88becdcc79db4d43c3d3a0a95e3e644a1bdc4126f4abb6e53bfa5c
+size 2146875

cyberpunk2077/clip_006/metallic.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:111d1ca85df1132c8aaf272c2cbe0b634f205f0a404be9705d2afedb99cf16fb
+size 4132344

cyberpunk2077/clip_006/normal.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8b749741e57e4fd22b79ec79597d802d1e27a4870c67d377a258244caba7797
+size 10882581

cyberpunk2077/clip_006/rgb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0e8f829c82ebd87862e50f5334621b876fc34f38d1f52ed7b091b6142e7bef1
+size 10434434

cyberpunk2077/clip_006/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a12384de3c149e7abaceaaf23568de7442e74b13dc086c1cd281bc56edcfb63
+size 11704729

cyberpunk2077/clip_008/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17af49607b71a34c4a93c57668a57bad1fb7f17059127f7a3d27022a6a1eddc7
+size 9310449

cyberpunk2077/clip_008/normal.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:440a01c1efae063370758dd0b65fee225529b10782c952655c9bd54e8f56906a
+size 13461891

cyberpunk2077/clip_008/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3362080063f0978d8cf8b312b1fada99ac0200ef775df0d11783e66298e70ec
+size 19446422

cyberpunk2077/clip_009/albedo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:898d6ee8ddd44951358992799e13a7eec6f8d0420571ff875f29118d03966248
+size 12090751

cyberpunk2077/clip_009/ambient_occlusion.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:130de58eeba1dc03b22fb9fb9759364abcc7121e2799255ad19ba93bb78e7116
+size 16123360

cyberpunk2077/clip_009/roughness.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16eda88e7daf3361db84997a89ea7104c9e2625b08a9bf4fa599d64702d1f846
+size 14296529

metadata.csv

@@ -0,0 +1,41 @@
+game,clip_id,source_session,source_clip,channels
+cyberpunk2077,clip_000,2025-12-08_19-51-54_cut,clip_006,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_001,2025-12-04_09-06-36,clip_233,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_002,2025-11-28_21-31-51,clip_000,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_003,2025-11-28_20-57-49,clip_000,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_004,2025-12-15_23-55-21,clip_041,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_005,2025-12-08_18-39-32_cut,clip_063,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_006,2025-11-28_19-57-23,clip_006,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_007,2025-12-08_21-00-08,clip_113,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_008,2025-12-08_21-55-09,clip_041,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_009,2025-12-11_11-30-22_cut,clip_007,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_010,2025-12-11_11-15-11,clip_018,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_011,2025-12-11_11-32-19,clip_008,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_012,2025-11-28_20-58-29,clip_000,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_013,2025-12-04_08-49-55,clip_063,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_014,2025-12-08_19-59-30,clip_047,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_015,2025-12-08_19-55-14,clip_035,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_016,2025-12-04_15-52-45_cut,clip_034,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_017,2025-12-08_15-01-08_cut,clip_134,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_018,2025-12-02_15-12-00,clip_149,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+cyberpunk2077,clip_019,2025-12-01_19-27-57,clip_109,albedo|ambient_occlusion|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_000,2026-01-02_23-10-42,clip_0010,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_001,2026-01-02_23-06-28,clip_0014,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_002,2025-12-28_16-49-53,clip_0000,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_003,2025-12-28_17-02-25,clip_0046,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_004,2025-12-28_18-15-03,clip_0016,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_005,2026-01-02_21-51-39,clip_0016,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_006,2025-12-28_16-14-06,clip_0022,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_007,2025-12-28_18-21-07,clip_0032,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_008,2026-01-02_22-56-37,clip_0006,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_009,2026-01-03_22-28-30,clip_0076,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_010,2025-12-28_16-52-40,clip_0020,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_011,2025-12-27_16-34-41,clip_0032,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_012,2026-01-02_22-46-42,clip_0025,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_013,2026-01-02_21-58-43,clip_0009,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_014,2025-12-28_16-42-22,clip_0023,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_015,2025-12-28_17-33-18,clip_0010,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_016,2025-12-28_18-10-29,clip_0000,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_017,2025-12-28_17-09-06,clip_0076,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_018,2026-01-02_22-41-18,clip_0020,albedo|depth|metallic|normal|rgb|roughness
+black_myth_wukong,clip_019,2025-12-28_17-30-51,clip_0007,albedo|depth|metallic|normal|rgb|roughness