Teach a model — from your own photos or sounds, in this tab.
Give a few examples per class (webcam, upload, or microphone), press Train, and a small model learns your task on your GPU — real transfer learning on a frozen MobileNetV3 backbone, nothing uploaded. A sound is turned into a spectrogram and goes through the same backbone. Then the head's real torch code is on the right. loading…
1 · Classes & examples
Hit a wall? The panel only does the preset task. Want a different head, another loss, a threshold? Edit the code and run it — same examples.
2 · The head — real torch code (edit me)
—
3 · Watch it learn · then point the camera / mic
loss
train a model, then turn on the webcam or microphone to predict live