A spiking neural network that segments satellite and space-debris trails in
wide-field astronomical images. Adjust the controls to see how thresholding and
Temporal Spike Coherence shape the final detections.
These are real model outputs, but the network does not run in your browser.
PyTorch has no WebAssembly build, so the probability maps were computed locally by the
model on validation frames from a different instrument than the one it trained on.
Everything below — thresholding, connected components, box matching, metrics — is
computed live here and matches the paper's evaluation code exactly. To segment your own
images, use the
model repository.