AutoGaze Official Demo

Attend Before Attention: Efficient and Scalable Video Understanding via Autoregressive Gazing

📄 Paper 🌐 Project Website

Upload Video or Image

File Info

Output FPS

Frames per second for displaying output videos (only affects playback speed)

Gazing Ratio

Max fraction of patches to gaze at per frame

0.01 1.35

Task Loss Requirement

Reconstruction loss threshold

0 1.5

Ready

Gazing %

# Gazed Patches

Total Patches

Original

Gazing Pattern (all scales)

Reconstruction

Gazing Pattern (individual scales)