Unveiling Graph Power: SegmentAnything and GCN Synergy for Instance Segmentation and Classification

In this paper, we propose a novel approach to improve instance segmentation and classification tasks by incorporating SegmentAnything [1] before using graph neural network frameworks. Before experimentation, we ensure comprehensive data labeling using SegmentAnything for instance segmentation and Grounding DINO [2] for class annotation. Our study aims to evaluate the effectiveness of this integration by conducting a comparative analysis with the state-of-The-Art Vision Transformer model [3]. Surprisingly, our experiments reveal that while the vision transformer has demonstrated remarkable performance in various tasks, it underperforms compared to our proposed approach on the same dataset. Our findings underscore the efficacy of integrating SegmentAnything with graph neural networks for instance segmentation and classification tasks, emphasizing the pivotal of those networks in advancing computer vision methodologies.