SAGAN: non-local attention and conditioning for adversarial synthesis

What changed

Adapted non-local self-attention to generator and discriminator representations, paired with spectral normalization on both networks and distinct update rates for long-range image synthesis.

Scientific scope

Adapted non-local self-attention to generator and discriminator representations, paired with spectral normalization on both networks and distinct update rates for long-range image synthesis. Self-attention, spectral normalization and two-timescale updates have credited predecessors. Attention maps and feature metrics do not prove object-level causal understanding or complete distribution coverage.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Self-attention, spectral normalization and two-timescale updates have credited predecessors. Attention maps and feature metrics do not prove object-level causal understanding or complete distribution coverage.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Self-Attention Generative Adversarial Networks

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction