we introduce SAVOIAS a visual complexity dataset that compromises of more than 1,400 images from seven image categories namely Scenes, Advertisements, Visualization and infographics, Objects, Interior design, Art, and Suprematism. The images in each category portray diverse characteristics including various low-level and high-level features, objects, backgrounds, textures and patterns, text, and graphics. The ground truth for SAVOIAS is obtained by crowdsourcing more than 37,000 pairwise comparisons of images using the forced-choice methodology and with more than 1,600 contributors using the Figure-Eight crowdplatform. The resulting relative scores are then converted to absolute visual complexity scores using the Bradley-Terry method and matrix completion. When applying five state-of-the-art algorithms to analyze the visual complexity of the images in the SAVOIAS dataset, we found that the scores obtained from these baseline tools only correlate well with crowd-sourced labels for abstract patterns in the Suprematism category (Pearson correlation r=0.84). For the other categories, in particular, the objects and advertisement categories, low correlation coefficients were revealed (r=0.3 and 0.56, respectively). These findings suggest that (1) state-of-the-art approaches are mostly insufficient and (2) SAVOIAS enables category-specific method development, which is likely to improve the impact of visual complexity analysis on specific application areas, including computer vision.
git clone the repositoty:
git clone https://github.com/esaraee/Savoias-Dataset.git
1420 images of the SAVOIAS dataset are located in the respective 7 categories the belong to in Images folder. In addition, if researchers are interested to find the original name of the images they can use the Name_Mapping folder to retreive those. The visual complexity ground-truth scores for the SAVOIAS images is located in Ground truth folder.
Sample images of the Savoias dataset and their corresponding energy maps from the fourth max-pooling layer in the VGG-16 architecture trained for the scene recognitiontask. The images from top left to bottom right belong to datasets Advertisement, Places2, MASSVIS (Visualization and Infographics), MSCOCO, IKEA, and art respectively. Notethat the last three images belong to the art dataset where the last sample is from the Suprematism category.
If you have any question, concern, or bug report, please file an issue in this repository's Issue Tracker and we will respond accordingly.
This research was partially funded by the following NSF Awards:
- NSF Award #1421943 RI: Small: Using Humans in the Loop to Collect High-quality Annotations from Images and Time-lapse Videos of Cells
- NSF Award #1838193 BIGDATA: IA: Multiplatform, Multilingual, and Multimodal Tools for Analyzing Public Communication in over 100 Languages
We would like to thank the students, Yifu Hu and Yi Zheng who prepared the images for the interior design category of our dataset.
Please cite the following papers in your publications if it helps you with your research.
@article{SaraeeJaBe18,
title={SAVOIAS: A Diverse, Multi-Category Visual Complexity Dataset},
author={Saraee, Elham and Jalal, Mona and Betke, Margrit},
journal={arXiv preprint arXiv:1810.01771},
year={2018}
}
@article{SaraeeJaBe20,
title={Visual complexity analysis using deep intermediate-layer features},
author={Saraee, Elham and Jalal, Mona and Betke, Margrit},
journal={Computer Vision and Image Understanding},
pages={102949},
year={2020},
publisher={Elsevier}
}
You can download the PDF of SAVAOIAS paper here and the PDF of our paper that uses deep learning on SAVOIAS dataset to predict the visual complexity here.
SAVOIAS dataset is freely and publicly available under Apache License 2.0. For further information regarding the copyright of the images used in the Savioas dataset, please check the license file.
Savoia is derived from Margherita di Savoia, a town in Italy. Here is a beautiful picture of Salt crystal in salt brine, salt production at Margherita di Savoia, Apulia, Italy by Franz Aberham via Getty Images.
