Datasets:
metadata
language:
- en
license: cdla-permissive-2.0
size_categories:
- 1K<n<10K
task_categories:
- object-detection
pretty_name: PubTables Dataset for Table Detection
dataset_info:
features:
- name: image
dtype: image
- name: objects
struct:
- name: bbox
sequence:
sequence: float64
- name: categories
dtype: string
splits:
- name: train
num_bytes: 228796393
num_examples: 1500
download_size: 224693659
dataset_size: 228796393
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
tags:
- Documents
- Tables
Dataset Details
Dataset Description
This dataset is a sampled version of bsmock/pubtables-1m consisting of 1500 images along with the PASCAL VOC formatted bounding boxes of multiple tables. It can be used for demonstration purposes in finetuning DETR and multimodal large language models.
Dataset Sources [optional]
- Repository: [bsmock/pubtables-1m] (https://huggingface.co/datasets/bsmock/pubtables-1m)
Uses
It can be used for finetuning DETR and TATR type models. Please check the original license in microsoft/table-transformer.