The Analyze capability is designed to explore model training results.

We took the VOC dataset and trained our own object detector.

Results can be seen in the Statistics page.

We can see the Confidence vs IoU review histogram for all the different categories or for each of them individually.

We can see the Confidence thresholds and IoU thresholds and how the affect the precision recall curve.

Everything here is interactive.

And most interestingly, we can see the confusion matrix, which is also interactive. So here we can see each misclassification or misdetection, and look at each of these images individually, where they come from from the entire data set. Plus, we can see all the metadata embedded on the image.

We can see the ground truth as well as the object output and we can see them with the confidence and all of the rest of the metadata in a visual form.