AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

HaloQuest: A Visual Hallucination Dataset for Advancing Multimodal Reasoning

Zhecan Wang, Garrett Bingham, Adams Wei Yu, Quoc V. Le, Thang Luong, Golnaz Ghiasi

ECCV 2024 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 22 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune BLIP2 (8B) on VQA v2 plus HaloQuest instead of VQA v2 only (ablation)BLIP2 (8B) · HaloQuest evalaccuracy (Auto-Eval): 12.3 -> 34.7Improved
fine-tune MiniGPT4 (13B) on VQA v2 plus HaloQuest (ablation)MiniGPT4 (13B) · HaloQuest evalaccuracy (Auto-Eval): 16.1 -> 40.1Improved
add HaloQuest training data (ablation)BLIP2 (8B) · VQA v2 (overall)exact match accuracy: 70.5 -> 71.0Improved
add HaloQuest training data (ablation)MiniGPT4 (13B) · VQA v2 (overall)exact match accuracy: 74.9 -> 74.9No clear effect
train MiniGPT4 on HaloQuestMiniGPT4 · POPE Randomaccuracy: 0.72 -> 0.80Improved
use advanced Langfun schema with main-point reasoning instead of text-only prompt (ablation)Auto-Eval · Auto-Eval agreement with human ratersagreement (%): 93.4 -> 95.3Improved
scale model sizezero-shot VLMs · HaloQuest eval (zero-shot)accuracy (Auto-Eval): BEiT-3 (0.7B) 40.0 vs LLaVA (13B) 10.9No clear effect
evaluate on generated images instead of real imagesbenchmark design · HaloQuest evalcorrelation across models (r): r=0.97No clear effect
use Auto-Eval instead of BLEU, CIDEr, ROUGE, METEOR as hallucination metricevaluation metric · HaloQuest evalcorrelation with human evaluation (r): r=0.99Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy