← All demos

Action Recognition

Recognizes one of 15 human actions from the current frame. It reads posture/context, not motion, and classifies the whole image (not per-person) — so on video it classifies each frame and smooths the label over time. Best with one person in frame.

🖼️ Image

Load an image

…or pick a sample:

🎬 Video

Load a video

…or pick a sample:

▶ Cycling ▶ Dancing

📷 Camera

▶ Start camera

Pick an image, play a video, or start your camera to begin.