Scenes-with-text Detection (v4.1)

About this version

  • Submitter: keighrim
  • Submission Time: 2024-03-07T03:29:41+00:00
  • Prebuilt Container Image:
  • Release Notes

    This version includes many bugfixes and clarification for the previous version.

    • more informative, consistent, and flask-friendly debug-level logging for future development
    • two additional pretrained models, including one based on convnext_tiny backbone for quicker annotation
    • TimePoint annotations is re-worked
      • classification-related props in TP anns are now all based on the “RAW” labels from classifier
      • all images classification results are now recorded as TP annotations regardless of TF annotations
    • added two runtime parameter
      • useStitcher - when "false", app will only generate TimePoint annotations, not stitching them into TimeFrame annotations
      • modelName - to pick a model between pre-built classifiers (by default, app will use the best performing model from training experiments)
    • updated to the latest mmif-python and clams-python, and thus no longer generates MMIFs with a non-existing version

About this app (See raw metadata.json)

Detects scenes with text, like slates, chyrons and credits.


(Note: “*” as a property value means that the property is required but can be any value.)

Configurable Parameters

(Note: Multivalued means the parameter can have one or more values.)

Name Description Type Multivalued Default Choices
startAt Number of milliseconds into the video to start processing integer N 0  
stopAt Number of milliseconds into the video to stop processing integer N 10000000  
sampleRate Milliseconds between sampled frames integer N 1000  
minFrameScore Minimum score for a still frame to be included in a TimeFrame number N 0.01  
minTimeframeScore Minimum score for a TimeFrame number N 0.5  
minFrameCount Minimum number of sampled frames required for a TimeFrame integer N 2  
modelName model name to use for classification string N 20240126-180026.convnext_lg.kfold_000 20240126-180026.convnext_lg.kfold_000, 20240212-132306.convnext_lg.kfold_000, 20240212-131937.convnext_tiny.kfold_000
useStitcher Use the stitcher after classifying the TimePoints boolean N true false, true
pretty The JSON body of the HTTP response will be re-formatted with 2-space indentation boolean N false false, true


(Note: “*” as a property value means that the property is required but can be any value.)

(Note: Not all output annotations are always generated.)