Fig. 3: Top weights of predicting emotion ratings from fc8 layer of AlexNet for six example emotions.
From: Object representations drive emotion schemas across a large and diverse set of daily-life scenes
The transformation matrix (1000 object labels x 20 emotions) between AlexNet’s fc8 layer and emotion ratings was averaged across 15 cross-validation folds. For six example emotions, the top ten positive and top ten negative weights (out of 1000 object labels) are displayed.
