Small interactive demos that run entirely in your browser. Most of them accompany a post on this blog.
-
LaFAM
Where does a vision model look when nobody tells it what to find? Label-free activation maps answer this for classifiers, self-supervised and CLIP models, on example images, your own pictures or your camera, all computed in your browser.
-
Region similarity
Which parts of two images look alike to a vision model? Lines connect the regions whose features match best, for every pair, the regions you select or the one under your pointer.
-
Latent space browser
UMAP maps of image features from supervised, self-supervised and CLIP models, pathology foundation models and StyleGAN encoders. Hover over the points to see the images behind them and select clusters to inspect them.
-
Python in the browser
A small Python console built on Pyodide and WebAssembly. NumPy and other packages load as soon as you import them.