Adapting SAM with Dynamic Similarity Graphs for Few-Shot Parameter-Efficient Small Dense Object Detection: A Case Study of Chickpea Pods in Field Conditions
This study adapts Segment Anything Model (SAM) for few-shot fine-tuning to detect small, dense objects like chickpea pods in complex agricultural settings. The method uses a Dynamic Similarity-based Graph Adaptation (DSGA) module combined with Low-Rank Adaptation (LoRA). This approach efficiently modifies SAM's image encoder using minimal trainable parameters, achieving superior segmentation performance even with limited training data.