Segment Annotation Workflow

This page explains the practical workflow for segment annotation in videos after anonymization.

Target Audience

  • Physicians documenting findings in recorded procedures

  • Clinical documentation staff

  • Trainers onboarding new users

Primary Component

  • frontend/src/components/VideoExamination/VideoExaminationAnnotation.vue

Clinical Purpose

  • Mark clinically relevant video intervals with labels.

  • Correct segment start/end boundaries.

  • Validate all segments before final use in downstream workflows.

How Users Enter This Workflow

  • From anonymization validation after approving a video.

  • Directly with a deep link/query that includes a video ID (video=<id>).

Preconditions

  • The selected video must have usable anonymized processed media.

  • Segment-validated videos remain selectable for review.

  • Segment editing is read-only after segment validation unless an explicit edit override is active.

Step-by-Step for Medical Users

  1. Select the video from the dropdown.

  2. Confirm timeline and labels are visible.

  3. Choose a label for the segment.

  4. Set label start at the current time.

  5. Set label end at the appropriate end time.

  6. Adjust boundaries by dragging/resizing if needed.

  7. Save local edits with Segmentänderungen speichern.

  8. Complete review with Alle Segmente validieren and confirm the dialog.

What Happens on Full Validation

  • The UI sends a bulk validation request for all current segments.

  • Current start/end boundaries are submitted.

  • The video is marked as validated in segment annotation context.

  • outside handling is included in this finalization step.

Operational Monitoring

The VideoExaminationAnnotation page has two expensive backend actions that are watchable at:

  • KI neu berechnen posts to /api/media/videos/<video_id>/segments/rerun-predictions/.

  • Außerhalb-Segmente schwärzen posts to /api/media/videos/<video_id>/segments/blacken-outside/.

These requests return 202 Accepted when the backend queued work. The follow-up Celery work does not run in the generic maintenance worker even though the task names are listed during that worker’s startup.

Prediction rerun

Prediction reruns dispatch endoreg_db.video_temporal_inference.

  • Celery queue: inference

  • Systemd service: lx-annotate-celery-inference-worker.service

  • Typical log terms: video_temporal_inference, rerun-predictions, predicting_segments, Task ... received, succeeded, failed

Monitor it with:

journalctl -u lx-annotate-celery-inference-worker.service -f -o short-iso

Segment censorship / outside blackening

Outside-segment censorship dispatches endoreg_db.video_post_validation_rebuild. This is intentionally routed to the frame extraction queue because the rebuild can perform staged media/frame regeneration.

  • Celery queue: frame_extraction

  • Systemd service: lx-annotate-celery-frame-extraction-worker.service

  • Typical log terms: video_post_validation_rebuild, blacken-outside, Starting staged frame extraction, Running FFmpeg, cleanup_, Task ... received, succeeded, failed

Monitor it with:

journalctl -u lx-annotate-celery-frame-extraction-worker.service -f -o short-iso

Combined operator view

Use this when testing both buttons from the UI:

journalctl \
  -u lx-annotate-boot.service \
  -u lx-annotate-celery-inference-worker.service \
  -u lx-annotate-celery-frame-extraction-worker.service \
  -f -o short-iso | rg -i 'rerun-predictions|blacken-outside|video_temporal_inference|video_post_validation_rebuild|Task .*received|succeeded|failed|Running FFmpeg|Starting staged frame extraction|cleanup_|predicting_segments'

lx-annotate-celery-worker.service is the maintenance/default worker. It may print endoreg_db.video_temporal_inference and endoreg_db.video_post_validation_rebuild in its startup task registry, but queue routing decides which worker consumes the task. Do not use it as the primary journal for these two buttons.

Training Notes

  • Teach users to finish boundary corrections before full validation.

  • Encourage short iterative saves before the final validation click.

  • Use keyboard shortcuts for efficient operation during longer videos.

Common Issues

No video can be selected

  • Cause: no video has usable anonymized media for the current workflow.

  • Action: return to anonymization workflow and check status.

Validation button reports no segments

  • Cause: no segments were created for the selected video.

  • Action: create at least one segment before full validation.

Segment boundaries look wrong after validation

  • Cause: stale local state before submit.

  • Action: reload segments, adjust boundaries, save, then validate again.

Technical References

  • frontend/src/components/VideoExamination/VideoExaminationAnnotation.vue

  • frontend/src/components/VideoExamination/Timeline.vue