The image segmentation feature integrates Meta AI's Segment Anything Model (SAM) to partition image pixels into distinct, meaningful segments. This state-of-the-art Vision Transformer based promptable segmentation algorithm with zero-shot generalization allows for precise isolation of image components.
GSense supports Vit-h, Vit-b SAM backbones and uses the hugging face transformers library. Downloading model checkpoints for default Vit-h and Vit-b weights is not required.
