treatment record
glitch-04
Pose skeletons and face mesh instead of segmentation contours, on the most recent 1,440 frames. The prompts say 'empty, unmanned, deserted', and the detector finds people anyway: 108 of 1,440 frames, 92 separate apparitions averaging 1.2 frames, mostly at 0.15-0.20 confidence. overlay hold 6 makes each one linger about a quarter second instead of a single frame.
- status
- test pass
- upstream
- pass-02 frames 4,692–6,131
- evidence
- A lossless frame sequence was written and manifested; the video is derived from it.
Its window on pass 2's full 43,200-frame axis.
what it looks like · 12 frames
f0047053:16
f0048253:21
f0049453:26
f0050893:32
f0052093:37
f0053293:42
f0054733:48
f0055933:53
f0057133:58
f0058574:04
f0059774:09
f0061214:15This treatment kept no sampled frames of its own, so these are the source frames it consumed — pass-02 4,692–6,131, before the treatment was applied.
No segmentation contours at all - a pose detections file carries keypoints but no masks, so --draw-mask has nothing to draw. This is the skeleton half of the look on its own, not a replacement for the contour baseline.
Upstream
- pass
- pass-02
- frames
- 4,692–6,131
- source
regenerable: hstack of passes/pass-02/{frames,text} 4692-6131 (image left, text right)- source sha256
cc24db21d5d0
Regenerating the source
ffmpeg -framerate 24 -start_number 4692 -i frames/frame_%06d.png -framerate 24 -start_number 4692 -i text/frame_%06d.png -frames:v 1440 -filter_complex '[0:v][1:v]hstack=inputs=2' -c:v prores_ks -profile:v 4 -pix_fmt yuv444p10le source.mov |
Render
- composite
- over
- device
- cpu
- frames
- 1,440
- clip length
- 1 m 00 s
- time to render it
- 20 m 56 s
- chunking
- 160-frame windows, 9 of them, 124-frame lead-in discarded
- overlay draws
- box, skeleton, face_mesh
- overlay hold
- 6 frames
- stroke / bracket / label floor
- 5 / 56 / 4,800 px²
Displacement layers
| grid | max offset | angle | sparsity | jitter | hold | stagger | seed | envelope |
|---|---|---|---|---|---|---|---|---|
| 4×3 | 32 | 90° | 0.5 | 0.8 | 8 | 0.6 | 5 | none |
| 8×7 | 22 | 0° | 0.8 | 0.8 | 6 | 0.5 | 9 | none |
| 20×18 | 14 | 0° | 0.93 | 0.8 | 4 | 0.8 | 13 | none |
What the detector found
- model
yolo11n-pose.pt- confidence floor
- 0.15
- detections
- 114
- per frame
- 0.079
| label | detections | |
|---|---|---|
| person | 114 |
Outputs
| file | frames | geometry | runtime | bytes | sha256 |
|---|---|---|---|---|---|
previews/glitch-04.skeleton-pops.f004692-006131.mp4 | 4,692–6,131 | 1536x1344 | 1 m 00 s | 297.2 MB | 93cf81345a9e |
Named and hashed, not hosted. Nothing in this repository is a master.
Frame sequence
- directory
frames/- files
- 1,440
- bytes
- 2.0 GB
- geometry
- 1536x1344
- lossless
- yes
- manifest
frames-manifest.json
The treatment's record, the same way a pass's frames are its record. preview.mp4 is derived from these, not the other way round — extracting frames back out of it returns the codec's opinion of the pixels.
Provenance
This treatment was run as a local process rather than dispatched through the mesh, so it has no task id, no lease and no event trail. That is a gap in the record, and it is recorded as one.