DSGCR: Decomposed Spectral Geometry-Aware Cross-Modal Semantic Representation for 3D Visual Grounding

3D visual grounding requires robust cross-modal representation to achieve fine-grained semantic alignment and precise geometric reasoning. However, most methods employ unimodal pre-trained encoders that transfer visual and linguistic knowledge independently, inducing domain shift and poor cross-modal alignment. Meanwhile, spatial modeling with handcrafted priors limits cross-modal geometric representation, struggling to capture complex object relations due to spectral bias. To address these challenges, we propose Text-Aware Feature Tuning (TFT) and Decomposed Spectral Geometry (DSG) to enhance cross-modal semantic representation. Specifically, TFT injects linguistic context into the visual hierarchy to mitigate domain shift and facilitate early cross-modal alignment. DSG employs a learnable Fourier basis and explicitly decomposes pairwise relations into symmetric and antisymmetric spectral components, allowing the model to capture high-frequency geometric details and direction-aware relations for precise spatial reasoning. Extensive experiments on ScanRefer, Nr3D and Sr3D validate the effectiveness of our method, demonstrating state-of-the-art performance with improvements of 2.05\% Acc@0.25 for 3DREC and 1.09\% mIoU for 3DRES on ScanRefer.

3D visual grounding aims to help computers understand a 3D scene and find the object described by a person in natural language, such as “the chair to the left of the table.” This ability is important for robots, augmented reality and assistive systems that need to interact with real-world environments. However, current methods often struggle because they learn visual and language information separately and have difficulty understanding detailed spatial relations between objects.In this work, we propose a method that helps the model better connect language descriptions with 3D visual information. One part of our method uses the text description to guide how the 3D scene is processed, so the model can focus on objects and attributes that are relevant to words. Another part improves the model’s ability to reason about spatial relations, such as direction and relative position, which are crucial for identifying the correct object in cluttered scenes. Experiments on three standard 3D visual grounding benchmarks show that our method improves both object localization and segmentation. These results suggest that better language-guided perception and spatial reasoning can make 3D scene understanding more accurate and reliable.