NGonthier/IconArt

🖼️ The dataset IconArt dataset was introduced in the following paper : "Weakly Supervised Object Detection in Artworks" Gonthier et al. ECCV 2018 Workshop Computer Vision for Art Analysis - VISART 2018. This datasest is designed to evaluate Weakly Supervised object detection methods in paintings.

You can also find project page for the paper here.

This dataset contains 5955 images (from WikiCommons) : a train set of 2978 images and a test set of 2977 images (for classification task). 1480 of the 2977 test images are annotated with bounding boxes for 10 visual categories. The classes are ‘angel’,‘beard’,‘capital’,‘Child_Jesus’,‘crucifixion_of_Jesus’,‘Mary’,‘nudity’,‘ruins’,‘Saint_Sebastien’,‘turban’.

Most of the methods only run evaluation on the easiest 7 classes : ‘angel’,‘Child_Jesus’,‘crucifixion_of_Jesus’,‘Mary’,‘nudity’, ‘ruins’,‘Saint_Sebastien’.

In this folder you can find 3 other folders, the JPEGImages one contains the JPEG images. The Annotations contain the bounding boxes in a PASCAL VOC template (XML file). The ImageSets/Main folder contain 3 files :

• train.txt : contain the name of the images of the train set
• test.txt : contain the name of the images annotated with an instance level
• IconArt_v2.csv the class information per image (0 or 1 per class) but also if the image below to the train set or test one and then if the image is associated to bounding boxes annotations (Anno column).

Exemples of prediction on test images

Predictions on the test images for a model trained in a weakly supervised way on the train set.

Reference

If you use IconArt please cite the related paper:

  @InProceedings{Gonthier_2018_ECCV_Workshops,
  author = {Gonthier, Nicolas and Gousseau, Yann and Ladjal, Said and Bonfait, Olivier},
  title = {Weakly Supervised Object Detection in Artworks},
  booktitle = {Proceedings of the European Conference on Computer Vision (ECCV) Workshops},
  month = {September},
  year = {2018}
  }