analiza aiqbench
This research presents CVIF, a training-free inference-time framework designed to enhance Multimodal Large Language Models' ability to interpret geometric diagrams by correcting reliance on textual priors. By locally reconstructing visual evidence and steering attention during semantic decoding, the method significantly improves accuracy on sparse cue datasets like PGPS9K and PGDP5K. This advancement addresses a core limitation in current MLLMs regarding precise spatial reasoning without requiring additional model training.
CVIF: A Criticality-Driven Visual Intervention Framework for Geometric Diagram Understanding in MLLMs
arXiv:2610.06399v1 Announce Type: new Abstract: Despite significant progress in visual tasks by Multimodal Large Language Models (MLLMs), geometric diagram understanding remains challenging due to the presence of sparse visual cues and ambiguous symbol-primitive associations. MLLMs may therefore rely on textual priors, producing interpretations that conflict with visual evidence. We introduce the training-free Criticality-Driven Visual Intervention Framework (CVIF), an inference-time method that localizes critical layers and executes visual interventions during the transition from evidence aggregation to semantic decoding. At these layers, a Geometry-Constrained Local Relation Reconstruction (GCLR) module selects and weights vertex-centered visual evidence, while an Adaptive Visual Steering Operator (AVSO) redistributes attention mass toward the selected tokens. Experiments on PGPS9K and PGDP5K show that CVIF raises Overall F1 from 77.85 to 85.58 and from 75.23 to 82.84, respectively, establishing a novel inference-time visual intervention paradigm.
Abstract:Despite significant progress in visual tasks by Multimodal Large Language Models (MLLMs), geometric diagram understanding remains challenging due to the presence of sparse visual cues and ambiguous symbol-primitive associations. MLLMs may therefore rely on textual priors, producing interpretations that conflict with visual evidence. We introduce the training-free Criticality-Driven Visual Intervention Framework (CVIF), an inference-time method that localizes critical layers and executes visual interventions during the transition from evidence aggregation to semantic decoding. At these layers, a Geometry-Constrained Local Relation Reconstruction (GCLR) module selects and weights vertex-centered visual evidence, while an Adaptive Visual Steering Operator (AVSO) redistributes attention mass toward the selected tokens. Experiments on PGPS9K and PGDP5K show that CVIF raises Overall F1 from 77.85 to 85.58 and from 75.23 to 82.84, respectively, establishing a novel inference-time visual intervention paradigm.
Artificial Intelligence (cs.AI)
arXiv:2610.06399 [cs.AI]
(or arXiv:2610.06399v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.06399 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 224 cuvinte
Subiecte conexe
Beyond Generation and Accuracy: Diagnosing and Enhancing Visual Chain-of-Thought for Geometry Problem Solving
arXiv · 14 sept. 2026
The Machines that Make the Machines
NVIDIA · 7 oct. 2026
Python package speeds up quantum protein folding by 30 percent
Quantum Zeitgeist · 2 sept. 2026
Automate legacy web applications with Amazon Bedrock AgentCore Browser Tool
AWS · 13 aug. 2026