This class was learned from 79 images of Souvla. The visualizations below show the maximally activated representations of what makes Souvla look like Souvla. Each image begins with a randomly-seeded noise image, which leads to a slightly different class representation. Below are 12 activations created from 12 random noise images.
- Images used for training: 79
- Image sources: Google (47) , Flickr (19)
- Bounding boxes and URLs: souvla.json