Line Detection

Both line detectors consume an image or crop and produce ImageCrop artefacts for text lines. Their only task command is run; neither exposes train.

LineDetectorRLSA

Detects printed-text lines using image processing, with optional deskewing and rule removal. It does not download model weights.

detect_lines:
  type: collectra.LineDetectorRLSA
  input: page
  output: lines
  deskew: true

Commands:

collectra --pipeline pipeline.yaml task detect_lines run images --output results
collectra --pipeline pipeline.yaml task detect_lines run --help

Configure algorithm parameters in YAML. See Quickstart for a complete pipeline, and ImageCrop to extract or evaluate the lines.

LineDetectorSurya

Uses the Surya text-detection model. Requires the Surya backend and its weights. See Licensing and optional backends for version-dependent code and model-weight terms.

detect_lines:
  type: collectra.LineDetectorSurya
  input: page
  output: lines
  merge_horizontal: false
  min_height: 0

Commands:

collectra --pipeline pipeline.yaml task detect_lines run images --output results
collectra --pipeline pipeline.yaml task detect_lines run --help

Both detectors accept the shared run options.