CHIRP: A Fine-Grained Benchmark for Open-Ended Response Evaluation in Vision-Language Models

TL;DR

This paper introduces CHIRP, a comprehensive benchmark for evaluating vision-language models' open-ended responses, addressing limitations of existing evaluation methods through a new long-form response benchmark and a novel VLM suite called Robin.

Contribution

The paper presents CHIRP, a new benchmark for robust VLM evaluation, and Robin, a novel suite of VLMs combining LLMs and vision encoders, to improve assessment methods.

Findings

Robin reveals shortcomings in current evaluation techniques.

CHIRP enables more thorough and reliable VLM assessment.

Open-source resources promote reproducibility and further research.

Abstract

The proliferation of Vision-Language Models (VLMs) in the past several years calls for rigorous and comprehensive evaluation methods and benchmarks. This work analyzes existing VLM evaluation techniques, including automated metrics, AI-based assessments, and human evaluations across diverse tasks. We first introduce Robin - a novel suite of VLMs that we built by combining Large Language Models (LLMs) and Vision Encoders (VEs) at multiple scales, and use Robin to identify shortcomings of current evaluation approaches across scales. Next, to overcome the identified limitations, we introduce CHIRP - a new long form response benchmark we developed for more robust and complete VLM evaluation. We provide open access to the Robin training code, model suite, and CHIRP benchmark to promote reproducibility and advance VLM research.