warping-machine

First, load an image model created in Teachable Machine. This only accepts zip files uploaded locally.

Next, load a project from Teachable Machine that contains different images than what the model was trained on, but labeled with the same class names.

Now you can visualize how well the model did on the generalization set.

Scan through individual classification errors, or compare how the images appear in MobileNet embedding space versus your model's embedding space.