@@ -180,9 +180,7 @@ We test the performance in two scenarios for each augmentation method: In the ZS
...
@@ -180,9 +180,7 @@ We test the performance in two scenarios for each augmentation method: In the ZS
\noindent All our results are generated following the procedure described in the experiments section. For the multiple labels approach three manually created labels per class are used. The automatic augmentation approach utilizes five labels: one manually created label and four augmented versions. In table \ref{tab:ZSL_GZSL} one can see the ZSL, seen and unseen accuracies, as well as the harmonic mean. Table \ref{tab:top1_top5} displays a more detailed view of the achieved unseen accuracies. It shows the top-1 and top-5 accuracies for our approaches with their standard deviations (std) over the eight splits.
\noindent All our results are generated following the procedure described in the experiments section. For the multiple labels approach three manually created labels per class are used. The automatic augmentation approach utilizes five labels: one manually created label and four augmented versions. In table \ref{tab:ZSL_GZSL} one can see the ZSL, seen and unseen accuracies, as well as the harmonic mean. Table \ref{tab:top1_top5} displays a more detailed view of the achieved unseen accuracies. It shows the top-1 and top-5 accuracies for our approaches with their standard deviations (std) over the eight splits.
Improvements on the ZSL accuracy, the unseen accuracy and the harmonic mean are achieved using the descriptive labels. The accuracies increase even further with the multiple labels approach.
Improvements on the ZSL accuracy, the unseen accuracy and the harmonic mean are achieved using the descriptive labels. The accuracies increase even further with the multiple labels approach. Using automatic augmentation performs worse compared to multiple manually created labels, but it still constitutes a relative 23\% increase over using only one descriptive label.
Using automatic augmentation performs worse compared to multiple manually created labels, but it still constitutes a relative 23\% increase over using only one descriptive label.
The seen accuracy stays within the same range, only experiencing a marginal increase for the two cases that use multiple labels. This behaviour along with a decrease in unseen accuracy is observed whenever batch normalization is applied to any of our approaches. Therefore it is only applied in the cases where multiple labels are used because they require batch normalization in order for the training to converge.
The seen accuracy stays within the same range, only experiencing a marginal increase for the two cases that use multiple labels. This behaviour along with a decrease in unseen accuracy is observed whenever batch normalization is applied to any of our approaches. Therefore it is only applied in the cases where multiple labels are used because they require batch normalization in order for the training to converge.
