Performance¶
WIP
This is a work in progress. refresh the page to see any recent changes made
t-SNE Visualization¶
t-SNE is a handy technique for visualizing complex datasets. It squashes high-dimensional data down into 2D or 3D so you can see it on a scatter plot. Basically, it maps similar objects to nearby points and different objects to distant ones.
Here’s an example of a t-SNE visualization from our UAV dataset. Since points that are close together are more similar, the clear separation you see here means our exemplars are high quality and well-defined.
Precision-Recall Curves¶
Precision-recall curves show you how well your model balances precision and recall at different thresholds. You’re looking for curves that hug the top-right corner—that means the model is doing a great job on both counts. The AP (Average Precision) number tells you the area under the curve for each class, which is a quick way to see which classes the model is nailing and which ones it's struggling with.
In this example, Mesodinium (AP=0.62) is doing okay, while Akashiwo (AP=1.00) is perfect.
Click on the image to see a larger version.
Confusion Matrix¶
A confusion matrix is the best way to see which classes your model is mixing up. Often, these mix-ups happen because of labeling errors, but sometimes it’s just because two classes look really similar and are hard to tell apart.
Click on the image to see a larger version.
🗓️ Updated: 2025-10-11