HumaniBench: A Human-Centric Framework for Large Multimodal Models Evaluation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Raza, Shaina, Narayanan, Aravind, Khazaie, Vahid Reza, Vayani, Ashmal, Radwan, Ahmed Y., Chettiar, Mukund S., Singh, Amandeep, Shah, Mubarak, Pandya, Deval
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917109230993408
author Raza, Shaina
Narayanan, Aravind
Khazaie, Vahid Reza
Vayani, Ashmal
Radwan, Ahmed Y.
Chettiar, Mukund S.
Singh, Amandeep
Shah, Mubarak
Pandya, Deval
author_facet Raza, Shaina
Narayanan, Aravind
Khazaie, Vahid Reza
Vayani, Ashmal
Radwan, Ahmed Y.
Chettiar, Mukund S.
Singh, Amandeep
Shah, Mubarak
Pandya, Deval
contents Although recent large multimodal models (LMMs) demonstrate impressive progress on vision language tasks, their alignment with human centered (HC) principles, such as fairness, ethics, inclusivity, empathy, and robustness; remains poorly understood. We present HumaniBench, a unified evaluation framework designed to characterize HC alignment across realistic, socially grounded visual contexts. HumaniBench contains 32,000 expert-verified image question pairs derived from real world news imagery and spanning seven evaluation tasks: scene understanding, instance identity, multiple-choice visual question answering (VQA), multilinguality, visual grounding, empathetic captioning, and image resilience testing. Each task is mapped to one or more HC principles through a principled operationalization of metrics covering accuracy, harmful content detection, hallucination and faithfulness, coherence, cross lingual quality, empathy, and robustness.We evaluate 15 state-of-the-art LMMs under this framework and observe consistent cross model trade offs: proprietary systems achieve the strongest performance on ethics, reasoning, and empathy, while open-source models exhibit superior visual grounding and resilience. All models, however, show persistent gaps in fairness and multilingual inclusivity. We further analyze the effect of inference-time techniques, finding that chain of thought prompting and test-time scaling yield 8 to 12 % improvements on several HC dimensions. HumaniBench provides a reproducible, extensible foundation for systematic HC evaluation of LMMs and enables fine-grained analysis of alignment trade-offs that are not captured by conventional multimodal benchmarks. https://vectorinstitute.github.io/humanibench/
format Preprint
id arxiv_https___arxiv_org_abs_2505_11454
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HumaniBench: A Human-Centric Framework for Large Multimodal Models Evaluation
Raza, Shaina
Narayanan, Aravind
Khazaie, Vahid Reza
Vayani, Ashmal
Radwan, Ahmed Y.
Chettiar, Mukund S.
Singh, Amandeep
Shah, Mubarak
Pandya, Deval
Computer Vision and Pattern Recognition
Artificial Intelligence
Although recent large multimodal models (LMMs) demonstrate impressive progress on vision language tasks, their alignment with human centered (HC) principles, such as fairness, ethics, inclusivity, empathy, and robustness; remains poorly understood. We present HumaniBench, a unified evaluation framework designed to characterize HC alignment across realistic, socially grounded visual contexts. HumaniBench contains 32,000 expert-verified image question pairs derived from real world news imagery and spanning seven evaluation tasks: scene understanding, instance identity, multiple-choice visual question answering (VQA), multilinguality, visual grounding, empathetic captioning, and image resilience testing. Each task is mapped to one or more HC principles through a principled operationalization of metrics covering accuracy, harmful content detection, hallucination and faithfulness, coherence, cross lingual quality, empathy, and robustness.We evaluate 15 state-of-the-art LMMs under this framework and observe consistent cross model trade offs: proprietary systems achieve the strongest performance on ethics, reasoning, and empathy, while open-source models exhibit superior visual grounding and resilience. All models, however, show persistent gaps in fairness and multilingual inclusivity. We further analyze the effect of inference-time techniques, finding that chain of thought prompting and test-time scaling yield 8 to 12 % improvements on several HC dimensions. HumaniBench provides a reproducible, extensible foundation for systematic HC evaluation of LMMs and enables fine-grained analysis of alignment trade-offs that are not captured by conventional multimodal benchmarks. https://vectorinstitute.github.io/humanibench/
title HumaniBench: A Human-Centric Framework for Large Multimodal Models Evaluation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2505.11454