An official website of the United States government
Recent advances in instruction tuning have led to the development of State-of-the-Art Large Multimodal Models (LMMs). Given the novelty of these models the impact of visual adversarial attacks on LMMs has not been thoroughly examined. We conduct a comprehensive study of the robustness of various LMMs against different adversarial attacks evaluated across tasks including image classification image captioning and Visual Question Answer (VQA). We find that in general LMMs are not robust to visual adversarial inputs. However our findings suggest that context provided to the model via prompts--such as questions in a QA pair--helps to mitigate the effects of visual adversarial inputs. Notably the LMMs evaluated demonstrated remarkable resilience to such attacks on the ScienceQA task with only an 8.10% drop in performance compared to their visual counterparts which dropped 99.73%. We also propose a new approach to real-world image classification which we term query decomposition. By incorporating existence queries into our input prompt we observe diminished attack effectiveness and improvements in image classification accuracy. This research highlights a previously under explored facet of LMM robustness and sets the stage for future work aimed at strengthening the resilience of multimodal systems in adversarial environments.
more »
« less
TurtleBench: A Visual Programming Benchmark in Turtle Geometry
Humans have the ability to reason about geometric patterns in images and scenes from a young age. However, developing large multimodal models (LMMs) capable of similar reasoning remains a challenge, highlighting the need for robust evaluation methods to assess these capabilities. We introduce TurtleBench, a benchmark designed to evaluate LMMs’ capacity to interpret geometric patterns—given visual examples, textual instructions, or both—and generate precise code outputs. Inspired by turtle geometry, a notion used to teach children foundational coding and geometric concepts, TurtleBench features tasks with patterned shapes that have underlying algorithmic logic. Our evaluation reveals that leading LMMs struggle significantly with these tasks, with GPT-4V achieving only 19% accuracy on the simplest tasks and few-shot prompting only marginally improves their performance (<2%). TurtleBench highlights the gap between human and AI performance in intuitive and visual geometrical understanding, setting the stage for future research in this area and stands as one of the few benchmarks to evaluate the integration of visual understanding and code generation capabilities in LMMs, setting the stage for future research.
more »
« less
- Award ID(s):
- 2046873
- PAR ID:
- 10635513
- Publisher / Repository:
- Association for Computational Linguistics
- Date Published:
- Page Range / eLocation ID:
- 12170 to 12188
- Format(s):
- Medium: X
- Location:
- Albuquerque, New Mexico
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Audio-visual representation learning aims to develop systems with human-like perception by utilizing correlation between auditory and visual information. However, current models often focus on a limited set of tasks, and generalization abilities of learned representations are unclear. To this end, we propose the AV-SUPERB benchmark that enables general-purpose evaluation of unimodal audio/visual and bimodal fusion representations on 7 datasets covering 5 audio-visual tasks in speech and audio processing. We evaluate 5 recent self-supervised models and show that none of these models generalize to all tasks, emphasizing the need for future study on improving universal model performance. In addition, we show that representations may be improved with intermediate-task fine-tuning and audio event classification with AudioSet serves as a strong intermediate task. We release our benchmark with evaluation code and a model submission platform to encourage further research in audio-visual learning.more » « less
-
Abstract The increasing integration of Visual Language Models (VLMs) into visualization systems demands a comprehensive understanding of their visual interpretation capabilities and constraints. While existing research has examined individual models, systematic comparisons of VLMs' visualization literacy remain unexplored. We bridge this gap through a rigorous, first‐of‐its‐kind evaluation of four leading VLMs (GPT‐4, Claude, Gemini, and Llama) using standardized assessments: the Visualization Literacy Assessment Test (VLAT) and Critical Thinking Assessment for Literacy in Visualizations (CALVI). Our methodology uniquely combines randomized trials with structured prompting techniques to control for order effects and response variability ‐ a critical consideration overlooked in many VLM evaluations. Our analysis reveals that while specific models demonstrate competence in basic chart interpretation (Claude achieving 67.9% accuracy on VLAT), all models exhibit substantial difficulties in identifying misleading visualization elements (maximum 30.0% accuracy on CALVI). We uncover distinct performance patterns: strong capabilities in interpreting conventional charts like line charts (76‐96% accuracy) and detecting hierarchical structures (80‐100% accuracy), but consistent difficulties with data‐dense visualizations involving multiple encodings (bubble charts: 18.6‐61.4%) and anomaly detection (25‐30% accuracy). Significantly, we observe distinct uncertainty management behavior across models, with Gemini displaying heightened caution (22.5% question omission) compared to others (7‐8%). These findings provide crucial insights for the visualization community by establishing reliable VLM evaluation benchmarks, identifying areas where current models fall short, and highlighting the need for targeted improvements in VLM architectures for visualization tasks. To promote reproducibility, encourage further research, and facilitate benchmarking of future VLMs, our complete evaluation framework, including code, prompts, and analysis scripts, is available athttps://github.com/washuvis/VisLit‐VLM‐Eval.more » « less
-
Our world offers a never-ending stream of visual stimuli, yet today’s vision systems only accurately recognize patterns within a few seconds. These systems understand the present, but fail to contextualize it in past or future events. In this paper, we study long-form video understanding. We introduce a framework for modeling long-form videos and develop evaluation protocols on large-scale datasets. We show that existing state-of-the-art short-term models are limited for long-form tasks. A novel object-centric transformer-based video recognition architecture performs significantly better on 7 diverse tasks. It also outperforms comparable state-of-the-art on the AVA dataset.more » « less
-
null (Ed.)Our world offers a never-ending stream of visual stimuli, yet today's vision systems only accurately recognize patterns within a few seconds. These systems understand the present, but fail to contextualize it in past or future events. In this paper, we study long-form video understanding. We introduce a framework for modeling long-form videos and develop evaluation protocols on large-scale datasets. We show that existing state-of-the-art short-term models are limited for long-form tasks. A novel object-centric transformer-based video recognition architecture performs significantly better on 7 diverse tasks. It also outperforms comparable state-of-the-art on the AVA dataset.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.18653/v1/2025.naacl-long.607
