stephenmcconnachie/0004-pdf-layout-detection

Layout detection with PP-DocLayout-L

Bounding-box layout predictions for images from stephenmcconnachie/0004-pdf-pages-test, produced by PaddleOCR's PP-DocLayout-L.

Processing details

• Source: stephenmcconnachie/0004-pdf-pages-test
• Model: PaddlePaddle/PP-DocLayout-L (~123M params (RT-DETR-L backbone))
• Samples: 50
• Processing time: 0.30 min
• Processing date: 2026-06-20 09:31 UTC
• Confidence threshold: 0.5
• Layout NMS: on
• Output column: layout (JSON-encoded list of detections)

Schema

Each row contains the original columns plus:

• layout: JSON string. List of detections:

  [
    {"bbox": [x1, y1, x2, y2], "label": "text", "score": 0.97, "cls_id": 2},
    {"bbox": [x1, y1, x2, y2], "label": "table", "score": 0.92, "cls_id": 5}
  ]
  

  Coordinates are in original input-image pixel space (top-left origin, [xmin, ymin, xmax, ymax]).
• inference_info: JSON list tracking every model that has been applied to this dataset (appended on each run).

Usage

import json
from datasets import load_dataset

ds = load_dataset("{output_dataset_id}", split="train")
detections = json.loads(ds[0]["layout"])
for det in detections:
    print(det["label"], det["score"], det["bbox"])

Reproduction

hf jobs uv run --flavor l4x1 -s HF_TOKEN \
    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/pp-doclayout.py \
    stephenmcconnachie/0004-pdf-pages-test <output> --model-name PP-DocLayout-L

Generated with UV Scripts.