Simple-PP-DocLayoutV3

Simple document layout analysis based on PP-DocLayoutV3 of PaddleOCR-VL-1.5 with fewer dependencies. Detect and classify text, titles, formulas, figures, and more in images. Works great together with Simple-GLM-OCR for Optical Character Recognition (OCR) for text extraction.

Example

fromsimplelayoutimportSimplePPDocLayoutV3model=SimplePPDocLayoutV3()
paths= ["testimage.png"]
results=model.run(paths)
importmatplotlib.pyplotasplt# plot resultsforpath, resultinzip(paths, results):
plt.figure(figsize=(7, 10))
plt.title(path)
image=result["image"]
plt.imshow(image)
crops= []
# plot detected objects in imageforbbox, outline, score, labelinzip(
result["boxes"],
result["outlines"],
result["scores"],
result["labels"]
):
# position of detection result in imagex0, y0, x1, y1=bboxplt.plot([x0, x1, x1, x0, x0], [y0, y0, y1, y1, y0])
# crop detected region for future usecrop=image.crop(bbox)
crops.append(crop)
# outline of result["mask"] (more precise than bbox for rotated text)x, y=outline.Tplt.fill(x, y, alpha=0.1, edgecolor=(0, 0, 0, 0.5))
# score describes certainty of detection resultstyle=dict(fc="blanchedalmond", ec="orange", alpha=0.8)
plt.text(x0, y0, f"{score:.2f}, {label}", bbox=style)
plt.show()
# show cropped regionsfori, cropinenumerate(crops[:48], 1):
plt.subplot(8, 6, i)
plt.imshow(crop)
plt.axis("off")
plt.tight_layout()
plt.show()

The detected regions can then be used for e.g. OCR with Simple-GLM-OCR.

fromsimplelayoutimportSimplePPDocLayoutV3fromsimpleglmocrimportSimpleGlmOcrlayout_model=SimplePPDocLayoutV3()
ocr_model=SimpleGlmOcr()
result, =layout_model.run(["testimage.png"])
forbbox, labelinzip(result["boxes"], result["labels"]):
crop=result["image"].crop(bbox)
text=ocr_model.run("Text Recognition:", crop)
print(f"{label}:\n\n{text}")
print("-"*80)

Installation

pip install torch numpy pillow matplotlib # matplotlib is optional, only used for example.py
git clone https://github.com/99991/Simple-PP-DocLayoutV3.git
cd Simple-PP-DocLayoutV3
python3 example.py