SAM is promptable 意味着模型可以输入多种 prompt 用于分割图像中的指定目标,且对于每个prompt 都会输出 3 个 mask in order to make SAM ambiguity-aware. 在模型设计角度,考虑到其使用的灵活性,SAM 由一个 image encoder、一个 prompt encoder 和一个 lightweight mask decoder 组成,其中 image encoder 只是...
Feature request Currently, exporting SAM models with optimum results in a single .onnx file (https://huggingface.co/Xenova/sam-vit-base/tree/main/onnx). It would be great if we could add an option to separate the encoder and decoder into...