Locate and segment objects.

Where is the pet, and which pixels belong to it?

Classifying an image as “pet” does not locate the animal. A box encloses a region; a semantic mask labels individual pixels. This lab lets you draw both and inspect a tiny learned pixel classifier.

Eight Oxford-IIIT Pet images are stored locally. The numeric asset is 181,192 bytes; no remote inference is used. Five official trainval images train the model. Three original test images are openly reused as development, so there is no untouched evaluation here.