Pith. sign in

REVIEW 6 cited by

ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features

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 2502.04320 v2 pith:47KEQGKB submitted 2025-02-06 cs.CV cs.LG

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

Do the rich representations of multi-modal diffusion transformers (DiTs) exhibit unique properties that enhance their interpretability? We introduce ConceptAttention, a novel method that leverages the expressive power of DiT attention layers to generate high-quality saliency maps that precisely locate textual concepts within images. Without requiring additional training, ConceptAttention repurposes the parameters of DiT attention layers to produce highly contextualized concept embeddings, contributing the major discovery that performing linear projections in the output space of DiT attention layers yields significantly sharper saliency maps compared to commonly used cross-attention maps. ConceptAttention even achieves state-of-the-art performance on zero-shot image segmentation benchmarks, outperforming 15 other zero-shot interpretability methods on the ImageNet-Segmentation dataset. ConceptAttention works for popular image models and even seamlessly generalizes to video generation. Our work contributes the first evidence that the representations of multi-modal DiTs are highly transferable to vision tasks like segmentation.

Discussion (0). Sign in to comment.

Forward citations

Cited by 6 Pith papers

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

  1. Text Template Tokens Are Implicit Semantic Registers in Diffusion Transformers

    cs.CV 2026-07 conditional novelty 7.0 of 10

    Chat-template tokens in LLM-conditioned DiTs act as implicit semantic registers: they absorb object identity from image latents and maintain it, while direct prompt-reading heads are causally inert.

  2. S3OD: Towards Generalizable Salient Object Detection with Synthetic Data

    cs.CV 2025-10 conditional novelty 7.0 of 10

    A 139k-image synthetic dataset with diffusion- and DINO-derived masks, trained with a multi-mask decoder, improves cross-dataset salient-object detection and reaches state-of-the-art after fine-tuning.

  3. LoRAShop: Training-Free Multi-Concept Image Generation and Editing with Rectified Flow Transformers

    cs.CV 2025-05 conditional novelty 7.0 of 10

    LoRAShop localizes each LoRA's effect to attention-derived spatial masks inside a Flux transformer, enabling training-free multi-concept image generation and editing.

  4. From Generation to Generalization: Emergent Few-Shot Learning in Video Diffusion Models

    cs.CV 2025-06 conditional novelty 6.0 of 10

    Frozen CogVideoX1.5, adapted with LoRA on 3 to 30 input-output videos, performs segmentation, pose estimation, and abstract reasoning (ARC-AGI 16.75%) with modest but real generalization.

  5. FAIR_XAI: Improving Multimodal Foundation Model Fairness via Explainability for Wellbeing Assessment

    cs.AI 2026-04 unverdicted novelty 5.0 of 10

    Zero-shot vision-language models are unreliable and vary widely for depression screening, and explainability-based fairness interventions often trade away accuracy without reliable fairness gains.

  6. 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.

Pith tools