dataset card
Browse files
README.md
CHANGED
|
@@ -7,11 +7,18 @@ hand-tracked, with synchronised IMU and per-frame hand pose.
|
|
| 7 |
|
| 8 |
---
|
| 9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
## What each episode contains
|
| 11 |
|
| 12 |
```
|
| 13 |
<run>/<seg>/
|
| 14 |
-
left.mp4 1920x1200 H.265, white-balanced,
|
| 15 |
right.mp4 the other eye of the stereo pair, same treatment
|
| 16 |
frames.csv frame_idx, ts_us <- the authoritative video clock
|
| 17 |
imu.csv ts_us, ax/ay/az (m/s2), gx/gy/gz (deg/s)
|
|
@@ -60,19 +67,13 @@ contain faces some detector saw at all, so faces no model detected are absent fr
|
|
| 60 |
numerator and the denominator. **Identifiable faces remain in this data.** Do not use it for face
|
| 61 |
recognition, re-identification, or any purpose requiring that individuals be unidentifiable.
|
| 62 |
|
| 63 |
-
##
|
| 64 |
-
|
| 65 |
-
Each eye is undistorted with **its own camera's calibration** (intrinsics + 8-parameter rational
|
| 66 |
-
distortion), not a shared one -- the units are the same 116-degree optics but their coefficients
|
| 67 |
-
differ enough that a shared correction would be wrong.
|
| 68 |
-
|
| 69 |
-
Calibrations report stereo RMSE around 0.2-0.7 px and pass a live 128-sample check. The correction
|
| 70 |
-
is zero at the optical centre and grows with radius, reaching roughly 12-96 px at the frame edge
|
| 71 |
-
depending on the unit. One caveat: RMSE only measures fit where the calibration target sat, so
|
| 72 |
-
far-corner geometry is extrapolated rather than verified.
|
| 73 |
|
| 74 |
-
|
| 75 |
-
|
|
|
|
|
|
|
|
|
|
| 76 |
|
| 77 |
## Hand pose
|
| 78 |
|
|
|
|
| 7 |
|
| 8 |
---
|
| 9 |
|
| 10 |
+
> **18 Aug 2026 — partial takedown, re-render in progress.** 64 of the original 149 episodes have
|
| 11 |
+
> been temporarily removed: they were rectified with a per-camera calibration whose undistortion
|
| 12 |
+
> map folds inside the frame, producing a visible circular seam. All 149 are being re-rendered
|
| 13 |
+
> without rectification and with a stricter face-blur setting, and will replace this set shortly.
|
| 14 |
+
> The 85 episodes currently here are unaffected by the seam.
|
| 15 |
+
|
| 16 |
+
|
| 17 |
## What each episode contains
|
| 18 |
|
| 19 |
```
|
| 20 |
<run>/<seg>/
|
| 21 |
+
left.mp4 1920x1200 H.265, white-balanced, faces pixelated (native lens geometry)
|
| 22 |
right.mp4 the other eye of the stereo pair, same treatment
|
| 23 |
frames.csv frame_idx, ts_us <- the authoritative video clock
|
| 24 |
imu.csv ts_us, ax/ay/az (m/s2), gx/gy/gz (deg/s)
|
|
|
|
| 67 |
numerator and the denominator. **Identifiable faces remain in this data.** Do not use it for face
|
| 68 |
recognition, re-identification, or any purpose requiring that individuals be unidentifiable.
|
| 69 |
|
| 70 |
+
## Lens geometry
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
|
| 72 |
+
Frames are delivered in the camera's native geometry -- **not rectified**. The 116-degree optics
|
| 73 |
+
are already close to rectilinear across most of the frame, and per-camera calibrations exist but
|
| 74 |
+
roughly half of them produce an undistortion map that folds back on itself away from the
|
| 75 |
+
calibration target, so applying them selectively would give an inconsistent dataset. Consistent
|
| 76 |
+
raw geometry was judged the better trade. Per-camera intrinsics are available on request.
|
| 77 |
|
| 78 |
## Hand pose
|
| 79 |
|