This work introduces XPlainVerse, a large-scale benchmark designed for joint deepfake detection and human-centered explanation, and proposes novel metrics, EntityScore and EvidenceScore, that measure reasoning fidelity by checking whether explanations correctly identify manipulated entities and visual evidence.
Abstract
As deepfake detection models increasingly produce natural language explanations, their reasoning often remains weakly grounded in visual artifacts, limiting reliability and user trust. Existing benchmarks mainly evaluate classification accuracy, overlooking whether explanations reflect the actual manipulations. This gap hinders progress toward deployable, explainable deepfake detection systems. To this end, we introduce XPlainVerse, a large-scale benchmark designed for joint deepfake detection and human-centered explanation. XPlainVerse comprises one million real and manipulated images, pairing authentic images from five established sources with forgeries generated by twelve off-the-shelf image editing and synthesis models. We further propose a multi-stage filtering pipeline, Edit-Check, to verify if manipulations satisfy their intended edits, enabling reliable reasoning supervision at scale. Beyond dataset scale, XPlainVerse provides two complementary explanation styles: technical explanations for expert analysis and simplified explanations optimized for non-technical users. To evaluate explanation quality beyond surface similarity, we propose novel metrics, EntityScore and EvidenceScore, that measure reasoning fidelity by checking whether explanations correctly identify manipulated entities and visual evidence. Human annotations on 2,000 explanation pairs validate our dataset quality against human judgment. We believe XPlainVerse will establish grounded explanation quality as a measurable dimension of deepfake detection and support scalable research on trustworthy, interpretable models.
The Explainable Deepfake Detection Challenge at ACM Multimedia 2026 is designed to benchmark this joint capability of classification metrics with semantic similarity, simplicity, and intent-aware grounding metrics that assess whether explanations identify the relevant manipulated entities and supporting visual evidence.
Abhijeet Narang, Kartik Kuckreja, Shreya Ghosh et al.· 1 citation
Feature-robust Augmentation is introduced, which comprises diversified degradation-aware augmentation strategies, and a supervised contrastive learning pattern paired with a mean-teacher architecture that stabilizes features against augmentations through consistency constraints that wins the first place in ACM Multimedia 2026 Explainable Deepfake Detection Challenge.
Zhu Xu, Jia-Qi Tang, Pokai Chen et al.· 0 citations
Although multimodal large language models (MLLMs) have shown substantial potential in visual understanding and graphic code generation, editing scientific figures through code presents a greater challenge: a model must jointly recover visual structure, ground the requested change, generate compilable code, and preserve all unrelated content. While existing TikZ benchmarks mainly focus on figure reconstruction and generation, few systematically evaluate instruction-guided scientific figure editing with compilable code. We introduce Edit2TikZ, a comprehensive benchmark for scientific figure editing tasks, featuring 1,548 diverse and high-quality samples. Edit2TikZ combines real-world and controlled synthetic edit cases, supports both textual and visual localization request, and contains multi-step editing, each with step-level annotations. We further construct a human-aligned evaluation framework to measure whether a requested edit is completed while irrelevant content is preserved. Utilizing Edit2TikZ, we evaluate 14 mainstream MLLMs and find that current systems remain unreliable: on average, proprietary models achieve a compilation success rate of merely 75% and remain limited in both figure restoration and edit correctness, while compact models below 9B struggle further with instruction following and complete figure generation. Therefore, we build a mixed training set TikZEditMix and adopt reconstruction-then-editing curriculum learning for compact models. On Qwen3.5-4B, this training improves the compilation success rate from 45.35% to 83.40% and yields an average improvement of 18.7 points across our proposed evaluation metrics. The code and data will be released at https://github.com/Solunny/Edit2TikZ.
Zongyun Zhang, Jiacheng Ruan, Xian Gao et al.· 0 citations
Concept Bottleneck Models (CBMs) are designed to make visual classification interpretable by expressing predictions through human-understandable concepts. Although interpretability is the central motivation for CBMs, they are still largely evaluated as predictive models by downstream classification accuracy, supplemented by isolated qualitative examples. This highlights a pressing need for quantitative measures, a challenge complicated by the infeasibility of ground-truth concept annotation at scale and the open nature of concept lists due to a lack of consensus. To fill this gap, we develop a multimodal large language model (MLLM) council that, given an image and its CBM explanation, produces an explanation quality score. To ground and validate the council, we first conduct a human study to establish a ground-truth reference for CBM explanation quality: for an image, annotators compare explanations from two of LF-CBM, VLG-CBM, and CBM-Suite and choose the more useful one, or mark them as equally good or equally bad, yielding 2700 judgments over 900 image-comparison items on CUB-200, ImageNet-100, and Places365. Against this human reference, our five-model council, consisting of open-weight MLLMs, recovers over 70% of strict human preference rankings, rising to 83% on items where human annotators unanimously agree. Building on this validated council, we introduce CBX-Bench, a public benchmark and leaderboard: authors of new CBMs can submit their model's explanations, and CBX-Bench scores them with the council and maintains dataset-level rankings of explanation quality. CBX-Bench thus provides a human-aligned, scalable evaluation of CBM explanations beyond accuracy and isolated qualitative examples. The benchmark is available at https://github.com/meric-karadag/cbx-bench.
Yusuf Meric Karadag, Gulay Oklan, Seref Baris Cagliyan et al.· 0 citations
Post-training quantization (PTQ) has become a practical solution for deploying deep learning models on resource-constrained edge devices by compressing high-precision floating-point weights into low-precision representations without requiring retraining. Past research has demonstrated that quantization largely preserves classification accuracy; however, whether it also preserves the model's internal reasoning remains an open question. This study presents a systematic evaluation on how static PTQ affects the interpretability / explainability of five widely used CNN architectures: VGG19, ResNet18, EfficientNet-B0, DenseNet161, and MobileNetV2 at INT8 and INT4 precision. We employ a dual interpretability framework that combines Grad-CAM for spatial attention analysis with LIME for input-level feature attribution, and systematically compare full-precision and quantized models on two binary classification datasets. Interpretability is evaluated using three complementary metrics: the Pearson correlation coefficient, structural similarity index, and top-20% IoU to capture distributional and structural variations in model explanations, supplemented by deletion/insertion faithfulness analysis. The results show that classification accuracy is not a reliable indicator of interpretability stability under reduced precision. DenseNet161 maintains strong feature consistency across both precision levels, whereas EfficientNet-B0, despite achieving competitive spatial attention and classification accuracy at INT8 precision, exhibits a substantial degradation in input-level feature attribution. These findings have direct implications for the trustworthy deployment of quantized models in applications with high interpretability requirements, demonstrating that architecture selection is as important as the quantization strategy.
Kazi Kamruzzaman Rabbi, Md. Zami Al Zunaed Farabe, M. S. Rahman· 0 citations
Visual Question Answering (VQA) systems have achieved impressive performance with the rise of large-scale vision–language models (VLMs). However, these models remain vulnerable to multiple forms of multimodal bias, severely limiting their robustness and generalization. Existing debiasing techniques mainly depend on post hoc evaluation or architectural modifications, while recent prompt-learning-based methods reveal new opportunities for aligning downstream tasks with pretrained models. In this work, we propose a unified prompt-driven debiasing framework that integrates generative prompt learning and a fuzzing-based bias correction mechanism. The generative prompt component reformulates VQA as a cloze-style masked prediction problem, leveraging pretrained language priors to improve semantic grounding. Meanwhile, the fuzzing-based module actively constructs unexpected test samples during training and employs a reflection mechanism to correct biased predictions in-loop, yielding inference-time robustness without additional test-time components. Extensive experiments on VQA-v2, VQA-CP, VQA-CE, GQA-OOD, and VQA-VS demonstrate that the proposed framework significantly improves both in-distribution (ID) accuracy and out-of-distribution (OOD) robustness, outperforming existing prompt-only or data-augmentation-only debiasing methods.
Yali Fan, Gangyu Huang, Qiwen Lu et al.· Multimodal Technologies and...· 0 citations
