3D-Aware Vision-Language Models Fine-Tuning with Geometric Distillation

2025-06-11

AI Summary

Vision-language models (VLMs) exhibit fundamental limitations in 3D spatial understanding. To address this, we propose an unsupervised, architecture-agnostic geometric knowledge distillation framework that implicitly injects geometric priors—such as sparse correspondences, relative depth estimates, and dense cost volumes—extracted from 3D foundation models (e.g., MASt3R, VGGT) into the pre-trained visual-language representation space of 2D VLMs. Our approach requires neither architectural modifications nor 3D ground-truth annotations. It jointly leverages contrastive distillation and cross-modal feature alignment to enhance 3D structural reasoning while preserving the VLM’s original multimodal capabilities. Extensive evaluation on multiple 3D vision-language reasoning and perception benchmarks demonstrates substantial improvements: spatial reasoning accuracy increases significantly, and computational overhead is reduced by over 40%. To our knowledge, this is the first method achieving lightweight, general-purpose, and efficient knowledge-transfer-based 3D enhancement for VLMs.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Vision-Language Models (VLMs) have shown remarkable performance on diverse visual and linguistic tasks, yet they remain fundamentally limited in their understanding of 3D spatial structures. We propose Geometric Distillation, a lightweight, annotation-free fine-tuning framework that injects human-inspired geometric cues into pretrained VLMs without modifying their architecture. By distilling (1) sparse correspondences, (2) relative depth relations, and (3) dense cost volumes from off-the-shelf 3D foundation models (e.g., MASt3R, VGGT), our method shapes representations to be geometry-aware while remaining compatible with natural image-text inputs. Through extensive evaluations on 3D vision-language reasoning and 3D perception benchmarks, our method consistently outperforms prior approaches, achieving improved 3D spatial reasoning with significantly lower computational cost. Our work demonstrates a scalable and efficient path to bridge 2D-trained VLMs with 3D understanding, opening up wider use in spatially grounded multimodal tasks.

Problem

• Enhancing 3D spatial understanding in Vision-Language Models
• Injecting geometric cues without architectural changes
• Improving 3D reasoning with lower computational cost

Innovation

• Lightweight fine-tuning with geometric distillation
• Injects 3D cues without architecture changes
• Uses sparse, depth, and dense 3D features