← all repositories

mit-han-lab/anycost-gan

A conditional GAN generator that produces consistent outputs across different computational budgets for fast interactive image editing.

anycost-gan
Velocity · 7d
+0.4
★ / day
Trend
steady
star history

Anycost GAN enables interactive image synthesis and editing by providing a single generator that can operate at varying computational costs through adaptive channel and resolution configurations. It employs sampling-based multi-resolution training, adaptive-channel training, and a generator-conditioned discriminator to maintain visual consistency between lightweight preview modes and full-quality outputs. The system allows users to preview edits at 5x faster speeds before finalizing high-quality results.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.