Is it concerning that there are short, sudden drops in prediction in the middle of a block otherwise solidly classified as rabbit/duck? I don't know much ML, does anyone know why it'd be so discontinuous?
When making the animation I didn't intend for the occlusion, but the fact that the occlusion causes the prediction to drop to zero is itself an interesting data point. Many objects in real life are occluded.
Like if something passes through your field of vision an you recognize it as lets say a god. But a second later you feel something was off, turn around, look at it in details and determine it to be a weird looking cat? Yeah, those "AI"s so far seem incapable of such deep thinking.