Dataset for "Visualizing image content to explain novel image discovery"

Visualizing Image Content to Explain Novel Image Discovery Jake Lee, Kiri Wagstaff Lee, J.H., Wagstaff, K.L. Visualizing image content to explain novel image discovery. Data Min Knowl Disc 34, 1777–1804 (2020). https://doi.org/10.1007/s10618-020-00700-0 This repository contains supplemental scripts and data used in the experiments presented in the paper. Data/ ImageNet-Random 'data/build-imagenet-random' provides the scripts necessary to compile the ImageNet-Random balanced and imbalanced subsets of the ILSVRC2012 training set. The ILSVRC2012 training set must be downloaded. 'random_classes.txt' contains the class definitions for the ImageNet-Random data set. 'python reprod_balanced.py' will reproduce the balanced variant of the ImageNet-Random data set used for experiments in the paper. 'python reprod_imbalanced' will reproduce the imbalanced variant. 'ilsvrc' defines the directory path of the ILSVRC2012 training set. 'output' defines the directory in which the data set will be compiled. 'filenames' defines the filepath of the exact images to be compiled. This should be 'balanced_filenames.txt' or 'imbalanced_filenames.txt'. 'python build_balanced.py' will compile the balanced variant of the ImageNet-Random data set using different images from the same classes. 'python build_imbalanced.py' will compile the imbalanced variant. The following parameters must be set within the script. 'ilsvrc' defines the directory path of the ILSVRC2012 training set. 'output' defines the directory in which the data set will be compiled. 'targets' defines the filepath of the class definitions. For ImageNet-Random, this should be 'random_classes.txt'. ---- ImageNet-Yellow 'data/build-imagenet-yellow' provides the scripts necessary to compile the ImageNet-Yellow subset of the ILSVRC2012 training set. The ILSVRC2012 training set must be downloaded. 'yellow_classes.txt' contains the class definitions for the ImageNet-Yellow data set. 'python reprod_yellow.py' will reproduce the ImageNet-Yellow data set used for experiments in the paper. The same parameters as above must be set within the script. 'filenames' should be set to 'yellow_filenames.txt' 'python build_yellow.py' will compile the ImageNet-Yellow data set using different images form the same classes. The same parameters as 'build_balanced.py' must be set within the script. 'target' should be set to 'yellow_casses.txt' ---- Mars-Curiosity 24 classes of the Mars-Curiosity data set can be accessed on Zenodo at https://zenodo.org/record/1049137. An additional class of 21 images, "sun", was added for our experiments, for a total of 25 classes with 6712 images. This additional class is included in 'data/mars-sun-class/' ---- STONEFLY9 The STONEFLY9 data set can be access at http://web.engr.oregonstate.edu/~tgd/bugid/stonefly9/. ---- Extracted features Features extracted from the data set are provided in the following directories: 'data/build-imagenet-random/balanced_feats' ''data/build-imagenet-random/imbalanced_feats' ''data/build-imagenet-yellow/yellow_feats' ''data/mars_feats' ''data/STONEFLY9_feats' Each directory contains three '.csv' files with extracted features. These features were extracted using the instructions described in the README of https://github.com/jakehlee/dmkd-vis-image-disc