Research

Paper

TESTING March 09, 2026

UNBOX: Unveiling Black-box visual models with Natural-language

Authors

Simone Carnemolla, Chiara Russo, Simone Palazzo, Quentin Bouniot, Daniela Giordano, Zeynep Akata, Matteo Pennisi, Concetto Spampinato

Abstract

Ensuring trustworthiness in open-world visual recognition requires models that are interpretable, fair, and robust to distribution shifts. Yet modern vision systems are increasingly deployed as proprietary black-box APIs, exposing only output probabilities and hiding architecture, parameters, gradients, and training data. This opacity prevents meaningful auditing, bias detection, and failure analysis. Existing explanation methods assume white- or gray-box access or knowledge of the training distribution, making them unusable in these real-world settings. We introduce UNBOX, a framework for class-wise model dissection under fully data-free, gradient-free, and backpropagation-free constraints. UNBOX leverages Large Language Models and text-to-image diffusion models to recast activation maximization as a purely semantic search driven by output probabilities. The method produces human-interpretable text descriptors that maximally activate each class, revealing the concepts a model has implicitly learned, the training distribution it reflects, and potential sources of bias. We evaluate UNBOX on ImageNet-1K, Waterbirds, and CelebA through semantic fidelity tests, visual-feature correlation analyses and slice-discovery auditing. Despite operating under the strictest black-box constraints, UNBOX performs competitively with state-of-the-art white-box interpretability methods. This demonstrates that meaningful insight into a model's internal reasoning can be recovered without any internal access, enabling more trustworthy and accountable visual recognition systems.

Metadata

arXiv ID: 2603.08639
Provider: ARXIV
Primary Category: cs.CV
Published: 2026-03-09
Fetched: 2026-03-10 05:43

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.08639v1</id>\n    <title>UNBOX: Unveiling Black-box visual models with Natural-language</title>\n    <updated>2026-03-09T17:16:39Z</updated>\n    <link href='https://arxiv.org/abs/2603.08639v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.08639v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Ensuring trustworthiness in open-world visual recognition requires models that are interpretable, fair, and robust to distribution shifts. Yet modern vision systems are increasingly deployed as proprietary black-box APIs, exposing only output probabilities and hiding architecture, parameters, gradients, and training data. This opacity prevents meaningful auditing, bias detection, and failure analysis. Existing explanation methods assume white- or gray-box access or knowledge of the training distribution, making them unusable in these real-world settings. We introduce UNBOX, a framework for class-wise model dissection under fully data-free, gradient-free, and backpropagation-free constraints. UNBOX leverages Large Language Models and text-to-image diffusion models to recast activation maximization as a purely semantic search driven by output probabilities. The method produces human-interpretable text descriptors that maximally activate each class, revealing the concepts a model has implicitly learned, the training distribution it reflects, and potential sources of bias. We evaluate UNBOX on ImageNet-1K, Waterbirds, and CelebA through semantic fidelity tests, visual-feature correlation analyses and slice-discovery auditing. Despite operating under the strictest black-box constraints, UNBOX performs competitively with state-of-the-art white-box interpretability methods. This demonstrates that meaningful insight into a model's internal reasoning can be recovered without any internal access, enabling more trustworthy and accountable visual recognition systems.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CV'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-09T17:16:39Z</published>\n    <arxiv:comment>Under review at IJCV</arxiv:comment>\n    <arxiv:primary_category term='cs.CV'/>\n    <author>\n      <name>Simone Carnemolla</name>\n    </author>\n    <author>\n      <name>Chiara Russo</name>\n    </author>\n    <author>\n      <name>Simone Palazzo</name>\n    </author>\n    <author>\n      <name>Quentin Bouniot</name>\n    </author>\n    <author>\n      <name>Daniela Giordano</name>\n    </author>\n    <author>\n      <name>Zeynep Akata</name>\n    </author>\n    <author>\n      <name>Matteo Pennisi</name>\n    </author>\n    <author>\n      <name>Concetto Spampinato</name>\n    </author>\n  </entry>"
}