WildVision: Evaluating Vision-Language Models in the Wild with Human Preferences

TL;DR

WildVision introduces WV-Arena and WV-Bench, innovative platforms for evaluating vision-language models with human preferences, revealing strengths and weaknesses of top models in real-world scenarios.

Contribution

The paper presents WildVision-Arena and WV-Bench, new benchmarks incorporating human preferences and GPT-4 judging, for comprehensive evaluation of vision-language models in real-world settings.

Findings

GPT-4V outperforms many models in recognition and reasoning tasks.

Top models still struggle with subtle cues, spatial reasoning, and hallucinations.

Current VLMs face safety and hallucination issues when provoked.

Abstract

Recent breakthroughs in vision-language models (VLMs) emphasize the necessity of benchmarking human preferences in real-world multimodal interactions. To address this gap, we launched WildVision-Arena (WV-Arena), an online platform that collects human preferences to evaluate VLMs. We curated WV-Bench by selecting 500 high-quality samples from 8,000 user submissions in WV-Arena. WV-Bench uses GPT-4 as the judge to compare each VLM with Claude-3-Sonnet, achieving a Spearman correlation of 0.94 with the WV-Arena Elo. This significantly outperforms other benchmarks like MMVet, MMMU, and MMStar. Our comprehensive analysis of 20K real-world interactions reveals important insights into the failure cases of top-performing VLMs. For example, we find that although GPT-4V surpasses many other models like Reka-Flash, Opus, and Yi-VL-Plus in simple visual recognition and reasoning tasks, it still faces challenges with subtle contextual cues, spatial reasoning, visual imagination, and expert domain knowledge. Additionally, current VLMs exhibit issues with hallucinations and safety when intentionally provoked. We are releasing our chat and feedback data to further advance research in the field of VLMs.