Fig. 2
From: Enhancing zero-shot scene recognition through semantic autoencoders and visual relation transfer
The motivation of the proposed visual relation transfer method, where the values on the line represent the distances between two different samples. Generally, the greater the distances between the seen and the unseen visual features, the greater the distances between the seen and the unseen semantic vectors. Note: three seen images and two unseen images in this figure are from the MIT67 dataset (https://web.mit.edu/torralba/www/indoor.html).
