CAPruner: Conceptual-Adjacent Scene Graph Pruner for Enhancing 3D Spatial Reasoning of Large Language Models
Researchers have developed CAPruner, a novel method for pruning scene graphs to enhance the 3D spatial reasoning capabilities of large language models. Existing pruning techniques often remove task-relevant information, but CAPruner integrates fuzzy semantic relevance with spatial proximity to identify and preserve critical relations. This approach, trained without costly relation-level annotations, significantly improves LLM performance on 3D vision-language tasks. AI
IMPACT Enhances LLM performance on 3D spatial reasoning tasks by optimizing scene graph processing.