PatFigCLS Dataset - Patent Figure Classification Dataset

Dataset Summary

The PatFigCLS dataset is introduced in the paper Patent Figure Classification using Large Vision-language Models accepted at ECIR 2025. The dataset is designed specifically for patent figure classification and evaluation across multiple aspects, including type, projection, objects and USPC class.

The PatFigCLS dataset is used alongside another dataset called PatFigVQA, which is intended for fine-tuning and evaluating Large Vision-language Models (LVLMs) in few-shot learning setting for patent figure visual question answering.

The dataset is sourced from two exisiting datasets:

Extended CLEF-IP 2011, and

DeepPatent2

Data Format

The dataset is stored in .tar files for fast and efficient read access.

Data Fields

 __key__: unique sample id

image.png: patent figure file

label.txt:  classification label

Data Splits

For each classification aspect, three data splits exist: `train_150`, `val` and `test`.

How to Use

The recommended approach is using the Python library `webdataset`. Below is an example code.

import io

from PIL import Image

from torchvision.transforms import Compose, ToTensor

import webdataset as wds

from braceexpand import braceexpand

def transform(image):

  return Compose([ToTensor()])(image)

dataset = (

  wds.WebDataset(

    braceexpand('PatFigCLS/object/train_150/shard-{000000..000042}.tar'),

    shardshuffle=1000

  )

  .shuffle(1000)

  .to_tuple('__key__', 'image.png', 'label.txt')

  .map_tuple(

    lambda key: key,

    lambda image: transform(Image.open(io.BytesIO(image))),

    lambda label: label.decode('utf-8'),

  )

)

dataloder = wds.WebLoader(dataset)

Source Code

The source code used to produce this dataset can be found at https://github.com/TIBHannover/patent-figure-classification

Licensing Information

PatFigCLS dataset is released under GNU General Public License v3.0.