Pick an image, then run. The first run downloads the four models (about 35 MB) and runs them in your browser; nothing is uploaded.

1 · Classification

What is in the image?
    output: 1000 class scores
    MobileNet v2 · ImageNet, 1,000 classes

    2 · Semantic segmentation

    Which class is each pixel?
    output: W × H × 21 per-pixel scores
    DeepLab v3 · PASCAL VOC, 20 classes + background

    3 · Object detection

    Where is each object, and what is it?
    output: M × (80 + 4)
    SSD Lite MobileNet v2 (COCO-SSD) · COCO, 80 classes

    4 · Instance segmentation

    Which pixels belong to each object?
    output: one W × H mask per person
    BodyPix (MobileNet v1) · people only, one mask per person