README.md

metadata

license: other
license_name: diogenes-evaluation-only
license_link: LICENSE
language:
  - en
tags:
  - video
  - gameplay
  - human-demonstration
  - keyboard-mouse
  - action-annotation
  - depth
  - camera-pose
  - multi-camera
  - world-model
  - non-trainable
size_categories:
  - n<1K
extra_gated_heading: Request access to the Diogenes Capability Demo v01
extra_gated_description: >-
  Access is restricted. Enter the access code supplied to you by Diogenes Lab.
  Requests submitted without a valid code are rejected. Nothing in this
  repository may be used to train, fine-tune, or evaluate any model.
extra_gated_button_content: Request access
extra_gated_fields:
  Access code: text
  Organization: text
  Intended use: text
  I confirm I will NOT use this data to train, fine-tune, or evaluate any model: checkbox

Diogenes Capability Demo v01

One place to see what the capture line produces, across every data type we currently collect, without downloading hundreds of gigabytes first.

Each video below is rendered directly from a real delivered package. The panels are the actual files — the action HUD is read from the frame record, the depth panel is the delivered EXR, the point cloud is back-projected from that depth using the delivered poses. Nothing is illustrative.

⚠️ Not for model training

Everything here is non_trainable and production_certified: false. Permitted use is evaluation of capture quality, schema, and time alignment. Not permitted: training, fine-tuning, or evaluating any model; redistribution outside approved collaborators.

Depicted game content remains the copyright of its respective publishers. Diogenes Lab claims no rights to it.

The videos

File	Shows	Length
video/A1-gameplay-action-record-en.mp4	Human gameplay with the per-frame action record: held keys, mouse buttons, pointer delta, decoded action labels, process audio, and the integrity counters — beside the frame they describe	24 s
video/A3-depth-fallout4-commonwealth-120s-en.mp4	A 120 s depth package: RGB, view-space linear depth in metres, and the fused world-space point cloud with the camera path	122 s

The four data types we collect

1. Human-annotator gameplay capture — the volume line

A person plays a commercial title on their own machine; we record the window, the input device stream, and the process audio on one clock.

Per clip: video.mp4 · frames.jsonl.zst · events.jsonl.zst · audio.pcm.zst · meta.json · manifest.json.

One row of frames.jsonl, one per video frame:

{"frame": 36, "session_frame": 26258, "video_t_us": 600000,
 "output_qpc_start": 62647900169, "output_qpc_end": 62648066836,
 "source_qpc": 62648006256, "source_seq": 12379,
 "duplicate": false, "source_superseded": 0, "queue_drop": 0,
 "keys": [87], "buttons": [], "actions": ["camera.yaw", "move.forward"],
 "pointerDx": 1, "pointerDy": 0, "pointerBins": [0, 0], "events": [], "src": "rec"}

Two things worth noticing. actions is a decoded label — the semantic meaning of the input under that title's binding, not just the scancode — so the record is portable across titles. And duplicate / source_superseded / queue_drop are shipped in the data rather than suppressed, so a buyer can measure our capture integrity instead of trusting it.

events.jsonl keeps the raw device stream with hardware timestamps alongside the frame-aligned view; the two are reconcilable, and neither is derived from the other after the fact.

Live sample: DiogenesLab/Diogenes_Gameplay_raw_sample_v01 and DiogenesLab/Diogenes_Gameplay_validation_sample_v01.

2. Depth + pose packages — the geometric line

Same capture, plus a per-frame linear depth map and a per-frame camera pose, so the footage can be lifted into 3D. Delivered as video + depth.zip (fp16 EXR) + action_camera_30fps.json + action_camera.json + systeminfo.json + QC/ + sha256.json.

Live sample: DiogenesLab/Diogenes_Depth_Pose_sample_v01, which includes the projection conventions needed to reproduce the reconstruction and the measured cross-frame consistency figure.

3. Engine-side multi-camera — the controlled line

Four cameras observing one scene on one clock, generated from a declared motion plan, with bidirectional image-space overlap measured per camera pair and gated at delivery. Delivered as four video lanes + four action_camera_N.json + covis.json + plan.json + provenance.json + QC.

spec/ in this repository carries the JSON Schemas for these packages. Sample footage is available under NDA rather than here, because the scene asset in the current validation packages is licensed for internal validation only.

4. Spatial derivatives

Fused point clouds, TSDF meshes (GLB/PLY), and self-contained interactive viewers built from the depth line. Produced on request per package.

Schemas

spec/ contains the delivery JSON Schemas as authored:

• action_camera_sfm_30fps.schema.json, action_camera_sfm_15fps.schema.json — depth line
• action_camera_mc_cam1.schema.json, action_camera_mc_camN.schema.json — multi-camera line
• plan.schema.json — the motion plan a controlled capture is generated from
• systeminfo.schema.json — capture surface geometry

Field naming in these schemas follows the requesting party's original samples, including some irregular key names. They are reproduced as-is on purpose; silently "correcting" a key name is how an integration breaks.

Titles

The client ships 15 titles; 8 are admitted for capture at present, spanning first-person shooters, third-person action RPGs, driving simulation, and real-time strategy. Per-title input coverage differs — some titles expose a richer key surface than others — and coverage is documented per title rather than averaged away.

Access

All Diogenes Lab sample repositories share one access code, but permission is granted per repository. Request each separately.