ilia
12e5d1aeb1
Consecutive captures whose printed PAGE X OF Y markers advance under the same total are grouped into one multi-page PDF, with duplicate pages keeping the better read. Naming gains doc-title/party components and a scan-date fallback so nothing lands in UNSORTED. Vision LLM default switches to minicpm-v (faster, fewer false blanks on dense mono pages).
86 lines
3.4 KiB
YAML
86 lines
3.4 KiB
YAML
# PaperPod tunable thresholds. All values here override the built-in defaults
|
|
# in paperpod/config.py; delete a key to fall back to the default.
|
|
|
|
capture:
|
|
# Frames per second to sample from the source video for analysis.
|
|
sample_fps: 8
|
|
# Frames are downscaled to this width for motion analysis (full resolution
|
|
# is still used for the final document crops).
|
|
processing_width: 960
|
|
|
|
audio:
|
|
# WAV sample rate for the extracted audio track (16 kHz mono = whisper-ready).
|
|
sample_rate: 16000
|
|
|
|
motion:
|
|
# Gaussian blur kernel applied before frame differencing (odd number).
|
|
blur_ksize: 21
|
|
# A pixel counts as "changed" if its gray level moved by more than this (0-255).
|
|
pixel_threshold: 12
|
|
# Fraction of changed pixels (0.0-1.0) above which a frame counts as "moving".
|
|
# Raise if sensor noise triggers false motion; lower if small gestures
|
|
# (page flips) are being missed.
|
|
threshold: 0.02
|
|
# A stable stretch must last at least this long (seconds) to count as a window.
|
|
stable_min_duration_s: 1.0
|
|
|
|
document:
|
|
# Contours smaller than this fraction of the frame area are ignored.
|
|
min_area_ratio: 0.04
|
|
# Contours larger than this fraction of the frame are rejected as likely
|
|
# background/lighting rather than a document.
|
|
max_area_ratio: 0.92
|
|
# Reject candidates whose long side is more than this many times their
|
|
# short side (filters out thin slivers/edge-strip false detections).
|
|
max_aspect: 6.0
|
|
# Canny edge detection thresholds.
|
|
canny_low: 50
|
|
canny_high: 150
|
|
# Contour search runs on a frame downscaled to this width (then the
|
|
# winning quad is scaled back up for the full-resolution crop). Keeps
|
|
# edge/gap-closing behavior consistent regardless of source resolution —
|
|
# important for 4K phone video.
|
|
detect_width: 1000
|
|
# Bake illumination-normalization + contrast enhancement into the saved
|
|
# crop/PDF (grayscale "flattened scan" look). OCR always uses this
|
|
# enhancement internally regardless of this setting.
|
|
enhance: false
|
|
# Auto-detect and correct 0/90/180/270 rotation before OCR/PDF export.
|
|
auto_rotate: true
|
|
# Expand the detected quad outward by this fraction before warping, to
|
|
# recover paper margins clipped by curl/wrinkles (straight-line quad vs.
|
|
# a slightly bowed real edge). refine.tighten below trims any extra
|
|
# mat/background this pulls in back off.
|
|
pad_margin: 0.04
|
|
# Tighten crops to the actual paper region (cuts mat margins/hands) and
|
|
# inpaint fingers pressing on the page.
|
|
refine: true
|
|
remove_fingers: true
|
|
# Split separate document placements when no contour is seen for this long.
|
|
event_gap_s: 2.0
|
|
# Ignore single-frame detection blips.
|
|
event_min_samples: 2
|
|
|
|
llm:
|
|
# Identify/name documents with a local Ollama vision model (all offline).
|
|
# Reads whole documents at once — far more robust than Tesseract for
|
|
# curled receipts and stylized logos — and flags quality issues
|
|
# (fingers_visible, blurry, glare) in export_summary.csv.
|
|
# minicpm-v: ~8-12s/doc; avoids qwen2.5vl's false "blank" on dense
|
|
# bank statements. OCR still fills date/total when the model hallucinates.
|
|
# Falls back to Tesseract when Ollama is down.
|
|
enabled: true
|
|
model: minicpm-v
|
|
base_url: http://localhost:11434
|
|
timeout_s: 180
|
|
|
|
speech:
|
|
# Window around a capture event in which to look for a spoken description.
|
|
window_before_s: 3.0
|
|
window_after_s: 5.0
|
|
|
|
output:
|
|
dir: ./output
|
|
# Paperless-ngx consume directory (module 7, not wired up yet).
|
|
consume_dir: null
|