Remove tooling and capture-hardware references from card
Browse files
README.md
CHANGED
|
@@ -42,21 +42,21 @@ extra_gated_fields:
|
|
| 42 |
|
| 43 |
# Diogenes CoD Sample v01
|
| 44 |
|
| 45 |
-
A small, inspection-oriented sample of PC gameplay
|
| 46 |
-
|
| 47 |
|
| 48 |
-
This release exists so collaborators can evaluate
|
| 49 |
time-alignment fidelity. It is a **format and quality sample, not a training corpus.**
|
| 50 |
|
| 51 |
## ⚠️ Not for model training
|
| 52 |
|
| 53 |
Every record in this dataset carries `non_trainable: true` and `production_certified: false`,
|
| 54 |
-
copied verbatim from the source
|
| 55 |
-
|
| 56 |
-
|
| 57 |
|
| 58 |
**本数据集不可用于任何模型训练。** 所有记录的 `non_trainable` 标记均为 `true`,
|
| 59 |
-
`production_certified` 均为 `false`,与源
|
| 60 |
未取得可训练所需的 consent grant、catalog 认证与 QA 流程。
|
| 61 |
|
| 62 |
Constraints that apply:
|
|
@@ -64,21 +64,21 @@ Constraints that apply:
|
|
| 64 |
- **No training, fine-tuning, or evaluation of models** on this data.
|
| 65 |
- The video depicts **Call of Duty®**, copyright Activision Publishing, Inc. Diogenes Lab claims
|
| 66 |
no rights to the depicted game content. Redistribution outside approved collaborators is not permitted.
|
| 67 |
-
-
|
| 68 |
|
| 69 |
-
Permitted: inspecting
|
| 70 |
time-alignment between video frames and input events.
|
| 71 |
|
| 72 |
## Dataset structure
|
| 73 |
|
| 74 |
-
Three parquet tables plus the
|
| 75 |
|
| 76 |
```
|
| 77 |
data/segments.parquet 18 rows one row per 10-second segment (index table)
|
| 78 |
data/frames.parquet 5,400 rows one row per video frame, with action labels
|
| 79 |
data/events.parquet 10,032 rows one row per raw input event
|
| 80 |
videos/<session_id>/<run_id>/<segment_id>.mp4
|
| 81 |
-
metadata/ catalog
|
| 82 |
metadata/sessions/ per-session session.json
|
| 83 |
```
|
| 84 |
|
|
@@ -91,9 +91,8 @@ metadata/sessions/ per-session session.json
|
|
| 91 |
| Duration | 180.0 s (3 min) |
|
| 92 |
| Frames | 5,400 @ 30 fps |
|
| 93 |
| Input events | 10,032 |
|
| 94 |
-
| Video | 345 MiB, H.264,
|
| 95 |
| Resolution | 1920×1080 |
|
| 96 |
-
| Source window | 2026-07-25 → 2026-07-26 |
|
| 97 |
|
| 98 |
Frames carrying at least one input signal: 4,057 / 5,400 (75.1%).
|
| 99 |
Duplicate frames: 2. Dropped frames: 0 — all 18 videos verify against their recorded SHA-256.
|
|
@@ -107,10 +106,10 @@ Duplicate frames: 2. Dropped frames: 0 — all 18 videos verify against their re
|
|
| 107 |
| `video_path` | string | repo-relative path to the mp4 |
|
| 108 |
| `video_bytes`, `video_sha256` | int64, string | integrity |
|
| 109 |
| `num_frames`, `num_events`, `duration_s` | int32, int32, float64 | volume |
|
| 110 |
-
| `width`, `height`, `fps` | int32, int32, float64 |
|
| 111 |
| `game_id`, `catalog_id`, `catalog_version`, `catalog_mode` | string | provenance |
|
| 112 |
| `mapping_id`, `mapping_version` | string | action-mapping identity |
|
| 113 |
-
| `
|
| 114 |
| `run_state`, `run_stop_reason` | string | `non_trainable_complete`, `foreground_lost` |
|
| 115 |
| `non_trainable`, `production_certified` | bool | compliance flags, verbatim |
|
| 116 |
|
|
@@ -126,8 +125,8 @@ One row per rendered frame, aligned to the video by `video_t_us`.
|
|
| 126 |
| `actions` | list[string] | semantic actions from the mapping (see below) |
|
| 127 |
| `pointer_dx`, `pointer_dy` | int32 | relative mouse delta for this frame |
|
| 128 |
| `pointer_bins` | list[int32] | quantized pointer delta under contract `dm-1` |
|
| 129 |
-
| `duplicate` | bool | frame repeated because no new source frame
|
| 130 |
-
| `queue_drop`, `source_superseded` | int32 |
|
| 131 |
| `num_frame_events`, `frame_events_json` | int32, string | events attributed to this frame |
|
| 132 |
|
| 133 |
Semantic actions present in this sample, by frame count:
|
|
@@ -147,9 +146,9 @@ Raw input stream at full device resolution — finer-grained than the per-frame
|
|
| 147 |
| field | type | description |
|
| 148 |
|---|---|---|
|
| 149 |
| `seq` | int64 | monotonic sequence within the run |
|
| 150 |
-
| `qpc` | int64 |
|
| 151 |
| `type` | string | `mouse_move`, `key_down`, `key_up`, `mouse_down`, `mouse_up` |
|
| 152 |
-
| `device` | string | salted per-device hash, e.g. `mouse:0b5c…` — not reversible
|
| 153 |
| `vk`, `scan_code`, `flags` | int32 | keyboard event detail |
|
| 154 |
| `dx`, `dy`, `absolute` | int32, int32, bool | pointer motion |
|
| 155 |
|
|
@@ -190,39 +189,35 @@ for r in sorted(rows, key=lambda r: r["frame"])[:5]:
|
|
| 190 |
print(r["frame"], r["video_t_us"], r["keys"], r["actions"], r["pointer_dx"], r["pointer_dy"])
|
| 191 |
```
|
| 192 |
|
| 193 |
-
##
|
| 194 |
|
| 195 |
-
|
| 196 |
-
|
| 197 |
-
|
| 198 |
|
| 199 |
-
Video frames and input events share
|
| 200 |
per-frame action attribution exact rather than interpolated. Segments are 10 s and independently
|
| 201 |
hashed. Runs terminate when the game window loses foreground (`foreground_lost`); a run that
|
| 202 |
-
detects any dropped
|
| 203 |
|
| 204 |
The sample is 6 contiguous segments from each of 3 sessions, so temporal continuity within a
|
| 205 |
session is preserved.
|
| 206 |
|
| 207 |
## Privacy
|
| 208 |
|
| 209 |
-
No audio, no webcam, no desktop
|
| 210 |
-
|
| 211 |
-
|
| 212 |
|
| 213 |
## Provenance
|
| 214 |
|
| 215 |
| | |
|
| 216 |
|---|---|
|
| 217 |
-
| Catalog | `
|
| 218 |
| Game profile | `call-of-duty-steam-release-test` v5 |
|
| 219 |
-
| Capture profile | `cod-steam-release-test-1920x1080-30-v1` |
|
| 220 |
-
| Device profile | `cod-steam-windows-nvidia-release-test-v1` |
|
| 221 |
| Mapping | `game.call-of-duty.steam.pc` v2026-07-24.2 |
|
| 222 |
| Input contract | `dm-1` (pointer_clamp 200, pointer_mu 8, pointer_max_bin 10) |
|
| 223 |
|
| 224 |
-
Full profile documents are reproduced verbatim under `metadata/`.
|
| 225 |
-
|
| 226 |
## Citation
|
| 227 |
|
| 228 |
Internal sample; not for citation. Contact Diogenes Lab for access questions.
|
|
|
|
| 42 |
|
| 43 |
# Diogenes CoD Sample v01
|
| 44 |
|
| 45 |
+
A small, inspection-oriented sample of PC gameplay recording: 1920×1080 @ 30 fps H.264 video
|
| 46 |
+
segments with **frame-aligned keyboard/mouse action annotations**.
|
| 47 |
|
| 48 |
+
This release exists so collaborators can evaluate recording quality, annotation schema, and
|
| 49 |
time-alignment fidelity. It is a **format and quality sample, not a training corpus.**
|
| 50 |
|
| 51 |
## ⚠️ Not for model training
|
| 52 |
|
| 53 |
Every record in this dataset carries `non_trainable: true` and `production_certified: false`,
|
| 54 |
+
copied verbatim from the source metadata. The originating catalog runs in `dev_non_trainable`
|
| 55 |
+
mode and the source game profile is named *"Call of Duty® Steam(release 测试,禁止训练)"* —
|
| 56 |
+
"release test, training prohibited".
|
| 57 |
|
| 58 |
**本数据集不可用于任何模型训练。** 所有记录的 `non_trainable` 标记均为 `true`,
|
| 59 |
+
`production_certified` 均为 `false`,与源元数据逐字一致。数据来源为 release 测试录制,
|
| 60 |
未取得可训练所需的 consent grant、catalog 认证与 QA 流程。
|
| 61 |
|
| 62 |
Constraints that apply:
|
|
|
|
| 64 |
- **No training, fine-tuning, or evaluation of models** on this data.
|
| 65 |
- The video depicts **Call of Duty®**, copyright Activision Publishing, Inc. Diogenes Lab claims
|
| 66 |
no rights to the depicted game content. Redistribution outside approved collaborators is not permitted.
|
| 67 |
+
- These runs predate the consent-grant pipeline; there is no `consent_id` attached.
|
| 68 |
|
| 69 |
+
Permitted: inspecting recording quality, validating the annotation schema, and reviewing
|
| 70 |
time-alignment between video frames and input events.
|
| 71 |
|
| 72 |
## Dataset structure
|
| 73 |
|
| 74 |
+
Three parquet tables plus the media, joinable on `(session_id, run_id, segment_id)`:
|
| 75 |
|
| 76 |
```
|
| 77 |
data/segments.parquet 18 rows one row per 10-second segment (index table)
|
| 78 |
data/frames.parquet 5,400 rows one row per video frame, with action labels
|
| 79 |
data/events.parquet 10,032 rows one row per raw input event
|
| 80 |
videos/<session_id>/<run_id>/<segment_id>.mp4
|
| 81 |
+
metadata/ catalog and profile documents, verbatim
|
| 82 |
metadata/sessions/ per-session session.json
|
| 83 |
```
|
| 84 |
|
|
|
|
| 91 |
| Duration | 180.0 s (3 min) |
|
| 92 |
| Frames | 5,400 @ 30 fps |
|
| 93 |
| Input events | 10,032 |
|
| 94 |
+
| Video | 345 MiB, H.264, ~16 Mb/s |
|
| 95 |
| Resolution | 1920×1080 |
|
|
|
|
| 96 |
|
| 97 |
Frames carrying at least one input signal: 4,057 / 5,400 (75.1%).
|
| 98 |
Duplicate frames: 2. Dropped frames: 0 — all 18 videos verify against their recorded SHA-256.
|
|
|
|
| 106 |
| `video_path` | string | repo-relative path to the mp4 |
|
| 107 |
| `video_bytes`, `video_sha256` | int64, string | integrity |
|
| 108 |
| `num_frames`, `num_events`, `duration_s` | int32, int32, float64 | volume |
|
| 109 |
+
| `width`, `height`, `fps` | int32, int32, float64 | video geometry |
|
| 110 |
| `game_id`, `catalog_id`, `catalog_version`, `catalog_mode` | string | provenance |
|
| 111 |
| `mapping_id`, `mapping_version` | string | action-mapping identity |
|
| 112 |
+
| `mean_bitrate_bps` | int64 | mean video bitrate |
|
| 113 |
| `run_state`, `run_stop_reason` | string | `non_trainable_complete`, `foreground_lost` |
|
| 114 |
| `non_trainable`, `production_certified` | bool | compliance flags, verbatim |
|
| 115 |
|
|
|
|
| 125 |
| `actions` | list[string] | semantic actions from the mapping (see below) |
|
| 126 |
| `pointer_dx`, `pointer_dy` | int32 | relative mouse delta for this frame |
|
| 127 |
| `pointer_bins` | list[int32] | quantized pointer delta under contract `dm-1` |
|
| 128 |
+
| `duplicate` | bool | frame repeated because no new source frame was available |
|
| 129 |
+
| `queue_drop`, `source_superseded` | int32 | pipeline health counters |
|
| 130 |
| `num_frame_events`, `frame_events_json` | int32, string | events attributed to this frame |
|
| 131 |
|
| 132 |
Semantic actions present in this sample, by frame count:
|
|
|
|
| 146 |
| field | type | description |
|
| 147 |
|---|---|---|
|
| 148 |
| `seq` | int64 | monotonic sequence within the run |
|
| 149 |
+
| `qpc` | int64 | high-resolution timebase tick (10 MHz) |
|
| 150 |
| `type` | string | `mouse_move`, `key_down`, `key_up`, `mouse_down`, `mouse_up` |
|
| 151 |
+
| `device` | string | salted per-device hash, e.g. `mouse:0b5c…` — not reversible |
|
| 152 |
| `vk`, `scan_code`, `flags` | int32 | keyboard event detail |
|
| 153 |
| `dx`, `dy`, `absolute` | int32, int32, bool | pointer motion |
|
| 154 |
|
|
|
|
| 189 |
print(r["frame"], r["video_t_us"], r["keys"], r["actions"], r["pointer_dx"], r["pointer_dy"])
|
| 190 |
```
|
| 191 |
|
| 192 |
+
## Recording methodology
|
| 193 |
|
| 194 |
+
Recording covers the game window only — no desktop, no other windows, no audio. Input is
|
| 195 |
+
captured at full device resolution in relative-pointer mode, restricted to the key and button
|
| 196 |
+
set declared in the game profile's input contract.
|
| 197 |
|
| 198 |
+
Video frames and input events share a single high-resolution timebase, which is what makes
|
| 199 |
per-frame action attribution exact rather than interpolated. Segments are 10 s and independently
|
| 200 |
hashed. Runs terminate when the game window loses foreground (`foreground_lost`); a run that
|
| 201 |
+
detects any dropped frame is failed and discarded wholesale rather than repaired.
|
| 202 |
|
| 203 |
The sample is 6 contiguous segments from each of 3 sessions, so temporal continuity within a
|
| 204 |
session is preserved.
|
| 205 |
|
| 206 |
## Privacy
|
| 207 |
|
| 208 |
+
No audio, no webcam, no desktop capture, and no keystrokes outside the game's declared input
|
| 209 |
+
contract. Device identifiers are salted hashes. The recording reflects a single internal
|
| 210 |
+
operator; no third-party personal data is present.
|
| 211 |
|
| 212 |
## Provenance
|
| 213 |
|
| 214 |
| | |
|
| 215 |
|---|---|
|
| 216 |
+
| Catalog mode | `dev_non_trainable` (v2026-07-24.5) |
|
| 217 |
| Game profile | `call-of-duty-steam-release-test` v5 |
|
|
|
|
|
|
|
| 218 |
| Mapping | `game.call-of-duty.steam.pc` v2026-07-24.2 |
|
| 219 |
| Input contract | `dm-1` (pointer_clamp 200, pointer_mu 8, pointer_max_bin 10) |
|
| 220 |
|
|
|
|
|
|
|
| 221 |
## Citation
|
| 222 |
|
| 223 |
Internal sample; not for citation. Contact Diogenes Lab for access questions.
|