Pith. sign in

REVIEW 10 cited by

What the DAAM: Interpreting Stable Diffusion Using Cross Attention

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2210.04885 v5 pith:PXHF6CJG submitted 2022-10-10 cs.CV cs.CL

classification cs.CVcs.CL
keywords daamdiffusionattributiongenerationqualitysemanticstableability
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Large-scale diffusion neural networks represent a substantial milestone in text-to-image generation, but they remain poorly understood, lacking interpretability analyses. In this paper, we perform a text-image attribution analysis on Stable Diffusion, a recently open-sourced model. To produce pixel-level attribution maps, we upscale and aggregate cross-attention word-pixel scores in the denoising subnetwork, naming our method DAAM. We evaluate its correctness by testing its semantic segmentation ability on nouns, as well as its generalized attribution quality on all parts of speech, rated by humans. We then apply DAAM to study the role of syntax in the pixel space, characterizing head--dependent heat map interaction patterns for ten common dependency relations. Finally, we study several semantic phenomena using DAAM, with a focus on feature entanglement, where we find that cohyponyms worsen generation quality and descriptive adjectives attend too broadly. To our knowledge, we are the first to interpret large diffusion models from a visuolinguistic perspective, which enables future lines of research. Our code is at https://github.com/castorini/daam.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 10 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features

    cs.CV 2025-02 conditional novelty 7.0 of 10

    ConceptAttention shows that linear projections in the output space of DiT attention layers yield sharper concept-localizing saliency maps than cross-attention maps, reaching state-of-the-art zero-shot segmentation.

  2. SAEmnesia: Erasing Concepts in Diffusion Models with Supervised Sparse Autoencoders

    cs.CV 2025-09 conditional novelty 6.0 of 10

    A supervised sparse autoencoder binds each concept to a single neuron, letting Stable Diffusion erase a concept by steering one latent.

  3. TARA: Token-Aware LoRA for Composable Personalization in Diffusion Models

    cs.CV 2025-08 conditional novelty 6.0 of 10

    TARA adds token-focused masking and a token alignment loss to LoRA adapters, allowing several independently trained personalized adapters to be composed with less identity loss and feature leakage.

  4. Heeding the Inner Voice: Aligning ControlNet Training via Intermediate Features Feedback

    cs.CV 2025-07 conditional novelty 6.0 of 10

    InnerControl trains lightweight probes on intermediate UNet features to enforce control alignment throughout the denoising trajectory, improving controllability for edges and depth.

  5. Diffusion Counterfactual Generation with Semantic Abduction

    cs.LG 2025-06 conditional novelty 6.0 of 10

    Diffusion-based causal image counterfactuals with semantic abduction improve identity preservation at a small cost in intervention effectiveness, demonstrated on Morpho-MNIST, CelebA-HQ, and mammogram artifact removal.

  6. Spatial Balancing: Designing an LLM-Powered Spatial Externalization Interface for Iterative Science Communication Writing

    cs.HC 2025-09 unverdicted novelty 5.0 of 10

    SpatialBalancing is a system that turns revision trade-offs into spatial navigation so writers can iteratively balance scientific exposition and narrative engagement with LLM assistance.

  7. Attention of a Kiss: Exploring Attention Maps in Video Diffusion for XAIxArts

    cs.AI 2025-08 conditional novelty 5.0 of 10

    A method and case study for visualizing cross-attention maps in Wan video diffusion transformers, showing token-region alignment over time and their use as artistic material.

  8. TimeMachine: Fine-Grained Facial Age Editing with Identity Preservation

    cs.CV 2025-08 unverdicted novelty 5.0 of 10

    TimeMachine proposes a diffusion model with age-aware cross-attention and a latent age classifier, plus a 1M-image HFFA dataset, claiming SOTA age editing with identity preservation.

  9. Unsupervised Class Generation to Expand Semantic Segmentation Datasets

    cs.CV 2025-01 conditional novelty 5.0 of 10

    A Stable Diffusion and SAM pipeline generates masked cutouts for novel classes, and mixing them into synthetic source images lets UDA segmentation models learn those classes without modifying the simulator or algorithm.

  10. Disentangling Granularity: An Implicit Inductive Bias in Factorized VAEs

    cs.LG 2025-05 reject novelty 3.0 of 10

    A V-shaped pattern in the training objective of factorized VAEs is attributed to a tunable 'disentangling granularity', but the effect is confounded because coarser granularity removes penalty terms by construction.

Pith tools