View a narrated version of the tutorial:
In this tutorial, we will examine how features can be visualized in Expresso. To obtain features, perform the steps suggested in Tutorial-6.
Click on Experiment View. On the screen that appears, click on button named Visualize deep network features .
From the drop-down menu titled Feature Sets, select the feature set named
If you wish to visualize the features on a per-image basis, select the option PER-IMAGE in the drop-down menu adjacent to Feature Sets. Alternately, if you wish to visualize the features on a per-layer basis, select the PER-LAYER option. In this tutorial, we will first visualize features on a per-layer basis. Select PER-LAYER option.
Select the layer from the rightmost listing. We will choose
pool1 layer. This lets us view
pool1 features for all images.
Select PER-IMAGE option from listing located at the right of Feature Selection listing. All the 5 features
(conv1, conv2, pool1, pool2, pool5) of a single image are visible in the visualization panel. The index of image can be specified using rightmost listing.
Sometimes, the visualizations may require additional transformations for improvement or to gain new perspective into the underlying data. These transformations can be applied in a pipeline fashion.
For example, select log from the list mentioned in bottom left.
Click on ‘=>’ button. Similarly add invert to the right list. Press Apply button below left listing to apply the pipeline of transform operations (log,invert) over the existing view.