sci-figure

Install the package with pip install -e ., then run sh-sci-fig paper.pdf -f 3 to extract Figure 3, or use --all to extract every figure. Output is saved as figure_3.png at 600 DPI by default.

Use the -s flag with the panel label, for example sh-sci-fig paper.pdf -f 2 -s c produces figure_2c.png. Sub-figure labels are detected via white-space grid analysis combined with Tesseract or EasyOCR recognition.

The tool prioritizes vector-based PDFs where captions and embedded images are machine-readable. Scanned PDFs may fail because caption detection relies on text extraction via pdfplumber, which requires a real text layer.

Splitting fails when OCR cannot recognize panel labels or no consecutive label sequence starting from (a) is found. In that case the tool falls back to returning the entire figure region with a warning instead of an incorrect crop.

Tesseract is the default OCR engine and must be installed separately. EasyOCR is available as an optional alternative via pip install sci-figure[ocr] and can be selected with --ocr easyocr, or OCR can be disabled with --ocr none.

Yes, use --render-page with --annotate to generate an annotated page image showing detected bounding boxes, then re-extract with --bbox x0,y0,x1,y1 and --page to specify exact pixel coordinates for the crop.