The authors propose D-Scope (Diffusion Scope) to connect SAE feature interpretation with generation control in diffusion transformers (DiTs).
Text queries retrieve SAE features for steering diffusion transformers
D-Scope matches target descriptions to visual centroids of activating patches, retrieving decoder directions that steer generation under spatially masked interventions.
Top University
Xinyue Xu · Jiahao Zhang · Lijie Hu · Peter Hase · Hao Wang
Pivotal Research · Mohamed bin Zayed University of Artificial Intelligence · Schmidt Sciences · Stanford University · University of Illinois at Urbana-Champaign
Research Digest··3 min read
The authors introduce D-Scope, a framework that retrieves sparse autoencoder (SAE) decoder directions from text queries by aggregating SigLIP 2 embeddings of highly activating image patches into visual centroids.
Why this paper
From Stanford University and 4 others
In one line
D-Scope retrieves sparse autoencoder decoder directions for diffusion transformers by matching text queries to visual centroids of activating patches, enabling interpretable generation steering.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§