Datasetpublic

onath/dlp-sample

IndicDLP Sample Sample subset of the IndicDLP (Indic Document Layout Parser) dataset. 23 languages: Assamese, Bengali, Bodo, Dogri, English, Gujarati, Hindi, Kannada, Kashmiri, Konkani, Malayalam, Maithili, Manipuri, Marathi, Nepali, Odia, Punjabi, Sanskrit, Sindhi, Santali, Tamil, Telugu, Urdu 12 document domains: Acts & Rules, Brochure, Form, Magazine, Manual, Newspaper, Notice, Novel, Question Paper, Research Paper, Syllabus, Textbook 42 layout classes (bounding box… See the full description on the dataset page: https://huggingface.co/datasets/onath/dlp-sample.

sourceHugging Facecc-by-4.0updated 6mo agoView on Hugging Face

0likes320downloads

Dataset Card

IndicDLP Sample

Sample subset of the IndicDLP (Indic Document Layout Parser) dataset.

—23 languages: Assamese, Bengali, Bodo, Dogri, English, Gujarati, Hindi, Kannada, Kashmiri, Konkani, Malayalam, Maithili, Manipuri, Marathi, Nepali, Odia, Punjabi, Sanskrit, Sindhi, Santali, Tamil, Telugu, Urdu
—12 document domains: Acts & Rules, Brochure, Form, Magazine, Manual, Newspaper, Notice, Novel, Question Paper, Research Paper, Syllabus, Textbook
—42 layout classes (bounding box annotations)
—Splits: train / validation / test

Configs (subsets)

Use the config dropdown in the viewer, or specify when loading:

Languages: assamese (Assamese), bengali (Bengali), bodo (Bodo), dogri (Dogri), english (English), gujarati (Gujarati), hindi (Hindi), kannada (Kannada), kashmiri (Kashmiri), konkani (Konkani), malayalam (Malayalam), marathi (Marathi), maithili (Maithili), manipuri (Manipuri), nepali (Nepali), odia (Odia), punjabi (Punjabi), sanskrit (Sanskrit), sindhi (Sindhi), santali (Santali), tamil (Tamil), telugu (Telugu), urdu (Urdu)

Domains: acts-and-rules (Acts & Rules), brochure (Brochure), form (Form), magazine (Magazine), manual (Manual), newspaper (Newspaper), notice (Notice), novel (Novel), question-paper (Question Paper), research-paper (Research Paper), syllabus (Syllabus), textbook (Textbook)

Usage

python

from datasets import load_dataset

ds = load_dataset("onath/dlp-sample")                          # all data
ds = load_dataset("onath/dlp-sample", "hindi")                 # Hindi only
ds = load_dataset("onath/dlp-sample", "form")                  # Forms only
ds = load_dataset("onath/dlp-sample", "hindi", split="train")  # Hindi train split

Annotation format

objects.bbox is in normalized YOLO format: [cx, cy, w, h] where values are in [0, 1].

Layout classes

advertisement, answer, author, chapter-title, contact-info, dateline, figure, figure-caption, first-level-question, flag, folio, footer, footnote, formula, header, headline, index, jumpline, options, ordered-list, page-number, paragraph, placeholder-text, quote, reference, second-level-question, section-title, sidebar, sub-headline, sub-ordered-list, sub-section-title, subsub-headline, subsub-ordered-list, subsub-section-title, subsub-unordered-list, sub-unordered-list, table, table-caption, table-of-contents, third-level-question, unordered-list, website-link