CLI and batch

ofiqpy

ofiqpy -i INPUT -o OUTPUT.csv

INPUT must exist and be a .jpg, .jpeg, .png, or .bmp file, or a directory containing at least one such file. Directories are searched recursively. Input discovery, config/model integrity verification, and model loading complete before the output file is created. Invalid or empty input exits nonzero.

python -m ofiqpy.batch

python -m ofiqpy.batch -i INPUT -o OUTPUT.csv [-w WORKERS] [--resume]

Flag	Meaning
-i	Existing image or recursively searched directory
-o	Canonical assessment CSV
-w	Worker processes; default 1
--resume	Skip exact full-path identities already in a valid output CSV

Every worker loads a complete model graph. Increasing -w can multiply memory use and is therefore an explicit operator choice. Multi-worker runs use Python's spawn process context so a threaded OpenCV/ONNX runtime is never inherited through fork.

Resume validates the complete canonical header before reading identities. Duplicate rows and truncated/widened rows in an existing output are rejected. Identity is the full supplied/discovered path, not the basename, so two recursive captures named capture.png remain distinct.

The first real assessment completes before the output file is opened. An unreadable image or preprocessing failure raises BatchAssessmentError and makes the command exit nonzero; no all-sentinel row is written for that item. If a later item fails, earlier flushed rows remain valid and can be continued with --resume.

CSV contract

The output is semicolon-delimited and uses the fixed canonical 28-component order:

Filename;UnifiedQualityScore;...;NoHeadCoverings;UnifiedQualityScore.scalar;...;NoHeadCoverings.scalar;assessment_time_in_ms;

• One native/raw and one scalar column per component.
• Successful scalar values are in [0, 100].
• FailureToAssess is encoded as raw 0.000000, scalar -1.
• Filename fields use standard CSV quoting, including paths containing semicolons.
• The header retains OFIQSampleApp's terminal empty field; data rows do not add one.

The layout matches the canonical OFIQ profile. It is not a general adapter for C++ configurations that select a smaller active measure map.