AIC CTU@AVerImaTeC: dual-retriever RAG for image-text fact checking

Herbert Ullrich; Jan Drchal

cs.CLFeb 16, 2026

AIC CTU@AVerImaTeC: dual-retriever RAG for image-text fact checking

Authors: Herbert Ullrich, Jan Drchal

Abstract

In this paper, we present our 3rd place system in the AVerImaTeC shared task, which combines our last year's retrieval-augmented generation (RAG) pipeline with a reverse image search (RIS) module. Despite its simplicity, our system delivers competitive performance with a single multimodal LLM call per fact-check at just $0.013 on average using GPT5.1 via OpenAI Batch API. Our system is also easy to reproduce and tweak, consisting of only three decoupled modules - a textual retrieval module based on similarity search, an image retrieval module based on API-accessed RIS, and a generation module using GPT5.1 - which is why we suggest it as an accesible starting point for further experimentation. We publish its code and prompts, as well as our vector stores and insights into the scheme's running costs and directions for further improvement.

Figures & tables

System	Question Score	Evidence Score	Verdict Accuracy	Justification Score
HUMANE	0.89	0.54	0.55	0.56
ADA-AGGR	0.37	0.46	0.54	0.43
AIC CTU (ours)	0.81	0.33	0.35	0.30
XxP	0.39	0.27	0.26	0.20
teamName	0.66	0.23	0.26	0.22
REVEAL	0.63	0.28	0.24	0.13

Table 1: System leaderboard showing performance metrics on AVerImaTeC test-split. Our system described in section 2 is highlighted with italics .

Evidence format	Question Score	Evidence Score	Verdict Accuracy	Justification Score
Answer only	0.86	0.27	0.31	0.28
Question + Answer	0.84	0.33	0.39	0.31
Declarative evidence	0.82	0.35	0.38	0.32

Table 2: Ablation study tweaking the evidence generation format from section 2 , iii. Scheme used in final submission is in italics.

Figure 3

Automated Fact-Checking

Automated fact-checking detects checkable claims, retrieves evidence, and assesses whether that evidence supports or refutes them.

See all 139 papers

FAGER: Factually Grounded Evaluation and Refinement of Text-to-Image Models
May 18, 2026Youngsun Lim, Cusuh Ham, Pin-Yu Chen +1Factual Consistency Evaluation Automated Evaluation
Boston University · Adobe · IBM Research
EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment
Oct 10, 2026ZiAn Wang, Ruiqi Liu, Zhengbo Zhang +10Automated Fact-Checking Tool-Using Vision-Language Agents
Jilin University · Institute of Automation, Chinese Academy of Sciences · Southwest University +6
AiSearch: Interactive Multi-Modal Search with VLMs
Oct 1, 2026Ali Koksal, Mei Chee Leong, Vicky Sintunata +2Multimodal IR Vision-Language Models
Institute of Advanced Intelligence and Computing (IAIC), Agency for Science, Technology and Research (A*STAR), Singapore

May 18, 2026J/K move · Enter open · S save

FAGER: Factually Grounded Evaluation and Refinement of Text-to-Image Models

Youngsun Lim, Cusuh Ham, Pin-Yu Chen, Deepti Ghadiyaram

Boston University · Adobe · IBM Research

Factual Consistency Evaluation Automated Evaluation T2I Generation Evaluation

Open paper AskPDF arXiv

Existing text-to-image (T2I) evaluation metrics mainly assess whether generated images align with information explicitly stated in the prompt, but often fail to capture factual requirements that are implicit, externally grounded, or identity-defining. As a result, they are not well suited for evaluating factual correctness in prompts involving scientific knowledge, historical facts, products, or culture-specific concepts. We propose FActually Grounded Evaluation and Refinement (FAGER), an agentic framework that evaluates whether generated images correctly reflect visually verifiable facts grounded in or implied by the prompt, while also providing actionable feedback for improvement. FAGER first constructs a structured factual rubric by combining LLM-based fact proposal with reference-guided visual fact extraction and verification, then converts the rubric into question-answer pairs for VLM-based evaluation. To validate FAGER as a factuality metric, we introduce a Factual A/B test, which measures whether a metric prefers factual reference images over corresponding generated images. Across five datasets spanning science, history, products, culture, and knowledge-intensive concepts, FAGER consistently outperforms prior metrics on this test. We further show that FAGER can be used to refine T2I outputs in a fully training-free manner, yielding substantial factuality gains across datasets.

Abstract

Automated Fact-Checking

Explore similar work

FAGER: Factually Grounded Evaluation and Refinement of Text-to-Image Models

EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment

AiSearch: Interactive Multi-Modal Search with VLMs