A proof of concept using TensorFlow and the imagenet_r dataset to classify images from the GVSU Art Gallery collection.
The ImageNet-R dataset is "a set of images labelled... by collecting art, cartoons, deviantart, graffiti, embroidery, graphics, origami, paintings, patterns, plastic objects, plush objects, sculptures, sketches, tattoos, toys, and video games". This makes it a good candidate dataset to demonstrate how image classification can work.
Below is the output of 5 images from the collection stored in the gallery/images directory. As you can see there are some classifications that work very well like "lakeside: 77.10%" with the Gulls of Leland painting. Others are less accurate classifications like the "oxcart" in Picnic at Macatawa.
make install
make run
Output
Predictions for autumn.jpeg:
valley: 50.67%
lakeside: 34.42%
castle: 3.70%
Predictions for camelia.jpeg:
overskirt: 31.56%
hoopskirt: 18.11%
groom: 12.17%
Predictions for gulls.jpg:
lakeside: 77.10%
breakwater: 12.34%
seashore: 3.71%
Predictions for picnic.jpg:
oxcart: 7.41%
cliff: 6.85%
wreck: 6.72%
Predictions for sunset.jpeg:
tray: 62.12%
ant: 5.22%
geyser: 4.66%