lancer1977/pdfparser repository overview

⁠Poppler PDF Watch Folder – organized segments + blank filtering

Many PDFs tile artwork into multiple image objects (quarters/strips) for rendering/compression. pdfimages extracts those objects exactly as stored, so you may see chopped segments.

This watcher supports:

• pages: render full pages to PNG (recommended for “what it looks like”)
• images: extract embedded image objects, then organize + filter
• both: do both

⁠Folder layout (under /volume1/docker/pdf mounted to /work)

• /work/in — drop PDFs here
• /work/out — outputs go here (one subfolder per PDF)
• /work/archive — processed PDFs are moved here
• /work/failed — PDFs that error or produce no outputs

⁠Output layout per PDF

/work/out/<pdf-stem>/

• pages/ — full rendered pages (MODE=pages or MODE=both)
• images/ — organized extracted assets (MODE=images or MODE=both)
  • p###/ — “good” assets grouped by page, renamed with metadata
  • _blank/ — mostly-white/empty images (backgrounds)
  • _tiny/ — small strips/icons/noise (below thresholds)
  • _masks/ — likely 1-bit masks/stencils
  • _unknown/ — anything that couldn’t be mapped cleanly
  • _list.txt — raw pdfimages -list output for debugging/tweaks
• .done.json — marker containing pdf hash + settings

⁠Filtering knobs (environment variables)

• MIN_W, MIN_H, MIN_PIXELS — bucket tiny stuff to _tiny
• BLANK_MAX_NONWHITE_PCT — bucket blank-ish to _blank (default 0.5%)

⁠Why this helps

• You keep the raw segments, but you’ll find them under images/p001, images/p002, etc.
• Junk (masks/tiny/blank) is separated so your main asset folders are usable.

⁠Deploy

docker build -t polyhydra/poppler-watcher:1.2 .
docker stack deploy -c stack.yml pdf