Object segmentation is a pivotal development in AI, enabling the precise identification and labeling of objects within images, which has significant implications across various fields such as autonomous vehicles, medical imaging, and surveillance. Traditionally, segmentation models required extensive annotated data, posing challenges due to the labor-intensive nature of image labeling. Encord addresses this by employing micro-models that automate the annotation process, reducing the need for exhaustive human input. The emergence of new foundational models like DINO, CLIP, and the Segment Anything Model (SAM) enhances the capabilities of these micro-models, allowing for advanced few-shot learning. SAM, developed by Meta AI, can generate segmentation masks based on input prompts without additional training, although it requires human prompting and class specification. The Personalized-SAM (Per-SAM) approach refines SAM by using target-guided attention and semantic prompting for more effective segmentation, as demonstrated in a benchmark comparison with the Mask R-CNN model using the DeepFashion-MultiModal dataset. Per-SAM outperforms traditional models, showing strong capability in few-shot learning by leveraging its advanced learning capacity, and highlighting the potential for further innovations in computer vision.