Salient Imagenet

Class: 265 (Toy Poodle), Feature: 203

This feature is core for class Toy Poodle. It is the fourth most important feature for class Toy Poodle.

Below, we display the five Toy Poodle images that most highly activate the feature, along with the Neural Activation Map (heatmap) and Feature Attack for each image. These are the same as the images shown to crowdworkers to annotate this class-feature pair.

We additionally provide the full responses of the five crowdworkers that annotated this pair, as well as links to other related pairs (i.e. same class or same feature).

MTurk Worker Annotations

core

spurious

core

spurious

Main Focus	Confidence	Explanation
Main Object	5	The main focus is the toy poodle, and in these images the main focus is right on the toy poodle, which is useful and relevant.
Main Object	4	The red region in the highlighted visual attributes focuses on the hair of a toy poodle dog.
Main Object	5	It's focusing on the eyes and ears.
Main Object	5	The red region mainly focuses on parts of the predicted object, such as it's ears or snout.
Main Object	4	Focus is on eyes and ears of the poodle.

core

Related Class-Feature Pairs

Important Features for Toy Poodle

There are 5 features annotated as core and 0 features annotated as spurious for Toy Poodle. Core features include 203, 218, 1059, 1745, 1985.

Classes where Feature 203 is among Most Important (i.e. top-5)

Feature 203 is core for the following classes: Shih Tzu, Dandie Dinmont Terrier, Lhasa Apso, Cocker Spaniel, Irish Water Spaniel, Toy Poodle, Miniature Poodle, Standard Poodle, Hair Wig.