shudder5305/pupilometer
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
1f8da0017c
pupilometer/GEMINI.md
Tempest 1f8da0017c feat: Integrate pupil segmentation infrastructure 
This commit introduces the necessary infrastructure for integrating pupil segmentation into the mono camera pipelines.

Key changes include:
- Modifying `gstreamer_pipeline.py` to add a tee element to split mono camera streams, creating a dedicated branch for segmentation output with a placeholder `videoconvert` element and `appsink`. This also includes new callbacks and data structures to handle the segmentation frames.
- Adding a new Flask route `/segmentation_feed/<int:stream_id>` to `app.py` to serve the segmentation video stream to the frontend.
- Updating `index.html` to display the new segmentation feed and implementing cache-busting for all video streams.
- Introducing `test_segmentation.py` to verify the functionality of the new segmentation feed.
- Refine existing UI and visual tests by updating locators and fixing indentation errors to accommodate the new segmentation feature and maintain test stability.
2025-12-11 17:52:08 +07:00

2.1 KiB
Raw Blame History

Pupil Segmentation Integration

  • Objective: Integrated Pupil segmentation into the mono camera pipelines.
  • Key Changes:
    • Modified src/unified_web_ui/gstreamer_pipeline.py to:
      • Add a tee element for mono camera streams to split the video feed.
      • Create a new branch for pupil segmentation with a videoconvert placeholder and a dedicated appsink (seg_sink_{i}).
      • Implement on_new_seg_sample_factory callback to handle segmentation data.
      • Added seg_frame_buffers and seg_buffer_locks for segmentation output.
      • Introduced get_seg_frame_by_id to retrieve segmentation frames.
      • Ensured unique naming for tee elements (t_{i}) in the GStreamer pipeline to prevent linking errors.
    • Modified src/unified_web_ui/app.py to:
      • Add a new Flask route /segmentation_feed/<int:stream_id> to serve the segmentation video stream.
      • Added datetime.utcnow to the Jinja2 context for cache-busting in templates.
    • Modified src/unified_web_web_ui/templates/index.html to:
      • Include a new "Segmentation Feed" section displaying the segmentation video streams, sourcing from /segmentation_feed/ with cache-busting timestamps.
      • Updated existing video feeds (video_feed) with cache-busting timestamps for consistency.
  • Testing:
    • Created tests/test_segmentation.py to verify the segmentation feed is visible and updating.
    • Updated src/unified_web_ui/tests/test_ui.py to refine locators (#camera .camera-streams-grid .camera-container-individual) for camera stream elements, resolving conflicts with segmentation feeds.
    • Updated src/unified_web_ui/tests/test_visual.py to refine locators (#camera .camera-mono-row, #camera .camera-color-row, #camera .camera-mono) to prevent strict mode violations and ensure accurate targeting of camera layout elements.
    • Fixed indentation errors in src/unified_web_ui/tests/test_visual.py.
  • Status: All tests are passing, and the infrastructure for pupil segmentation is in place, awaiting the integration of a DeepStream model.