cs.CVNov 1, 2026

Vocabulary Aligned Explanations for Computer Vision Models

Authors: Erfan Mirhaji, Nikhil Krishnaswamy, Jill Zarestky, Lisa Mason, Nathaniel Blanchard, Sarath Sreedharan

Abstract

Optimizing vision models for accuracy does not force the model to learn and utilize human-understandable and domain-specific concepts and rules. We propose explaining image classifications through domain-specific concepts, visual evidence, and expert knowledge expressed as logical rules. We introduce FLUENT (Framework for Localized, Uncertainty-Aware Explanations in Normative Terminology), a post-hoc framework that ranks expert-authored rules combining jointly required, alternative, and absent concepts. To account for prediction errors, FLUENT combines predicted labels and concept states with empirically estimated likelihoods and concept priors. It scores each rule by calculating support over satisfying assignments to the concepts mentioned in that rule, then normalizes these scores to express relative confidence among candidate explanations. The selected rule is illustrated through occlusion-based highlights for positive concept mentions in the query image and localized examples in retrieved reference images for negative mentions. These illustrations are intended to show end users unfamiliar with the domain what the concepts named in an explanation look like. FLUENT combines domain knowledge and model accuracy, and delivers visual concept highlights and explicit rule confidence to explain an image in the vocabulary of an application domain.

Figures & tables

Appendix figures & tables26 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Mar 13, 2025

VisionLogic: Discovering and Grounding Decision-Relevant Visual Concepts

Concept-based explanations help users understand vision models through recognizable visual patterns. However, existing methods often rely on correlational signals without directly validating which image cues support prediction-relevant internal features. To this end, we introduce VisionLogic, a post-hoc framework that grounds these features in visual concepts through intervention-based validation. VisionLogic first identifies compact sets of features whose contributions reproduce the model's original prediction. It then represents their activation states as predicates using class-specific thresholds. An iterative refinement procedure grounds these predicates in visual regions through ablation tests. A region is accepted when its removal deactivates the corresponding predicate, linking the feature's numerical role to visual evidence in the input. The same predicates allow us to examine how features are activated, selected, and reused across images and classes. Across CNNs and vision transformers on ImageNet-1k, we find that only a few features are selected to explain each prediction, and frequently active features are not always selected. In a large-scale human evaluation with 465 participants, VisionLogic significantly improves participants' understanding of model behavior over established methods ACE and CRAFT. Code is available at https://github.com/allengeng123/VisionLogic.
May 7, 2026

Concept-Based Abductive and Contrastive Explanations for Behaviors of Vision Models

Concept-based explanations offer a promising approach for explaining the predictions of deep neural networks in terms of high-level, human-understandable concepts. However, existing methods either do not establish a causal connection between the concepts and model predictions or are limited in expressivity and only able to infer causal explanations involving single concepts. At the same time, the parallel line of work on formal abductive and contrastive explanations computes the minimal set of input features causally relevant for model outcomes but only considers low-level features such as pixels. Merging these two threads, in this work, we propose the notion of concept-based abductive and contrastive explanations that capture the minimal sets of high-level concepts causally relevant for model outcomes. We then present a family of algorithms that enumerate all minimal explanations while using concept erasure procedures to establish causal relationships. By appropriately aggregating such explanations, we are not only able to understand model predictions on individual images but also on collections of images where the model exhibits a user-specified, common behavior. We evaluate our approach on multiple models, datasets, and behaviors, and demonstrate its effectiveness in computing helpful, user-friendly explanations.
Jul 8, 2026

Naming the Concepts Classifiers Rely On: Language-Anchored Decomposition for Faithful Explanation

Deep neural networks are widely deployed in high-stakes visual applications where interpretability is critical, yet existing explanations face a trade-off: post-hoc concept methods recover factors that are faithful to a model's behavior but unnamed, while naming and by-design methods attach human-readable concepts only by retraining or altering the classifier. We propose Language-Anchored Decomposition (LAD), a post-hoc framework that delivers concepts which are simultaneously named, faithful, and obtained without modifying the model. For each class, a large language model proposes a concept vocabulary that CLIP-based similarity maps localize across image regions. Inverting standard non-negative matrix factorization, LAD fixes these language-grounded maps as the coefficient matrix and learns only a concept basis that reconstructs the frozen encoder's activations, so naming becomes a structural constraint and the model's own feature geometry determines which concepts are retained. Removing this anchor preserves accuracy but collapses attribution faithfulness. Across natural-image, scene, and medical-imaging benchmarks, LAD produces spatially precise explanations that are decision-relevant under both concept insertion and deletion, while uniquely providing stable, human-interpretable concept names.