How to store embedding and then use it in the SamPredictor Class?

[StraigenDaigen]

I would like to calculate embedding in a machine, then use it in another to do predictions. I have been trying to save it using

predictor = SamPredictor(sam) predictor.set_image(img) embedding = predictor.get_image_embedding().detach().cpu().numpy() with open(pickle_file_name, "wb") as f: pickle.dump(embedding, f)

then I upload the pickle file in another script but I don't know how to insert embedding again in the predictor avoiding to calculate embedding again with set_image(img).

[Ellen7ions]

I think u can do it in a simple way like this. :)

class SamPredictor:

    def save_image_embedding(self, path):
        if not self.is_image_set:
            raise RuntimeError("An image must be set with .set_image(...) before embedding saving.")
        res = {
            'original_size': self.original_size,
            'input_size': self.input_size,
            'features': self.features,
            'is_image_set': True,
        }
        torch.save(res, path)

    def load_image_embedding(self, path):
        res = torch.load(path, self.device)
        for k, v in res.items():
            setattr(self, k, v)

[qq846511277]

我认为你可以用这样简单的方式做到这一点。:)

class SamPredictor:

    def save_image_embedding(self, path):
        if not self.is_image_set:
            raise RuntimeError("An image must be set with .set_image(...) before embedding saving.")
        res = {
            'original_size': self.original_size,
            'input_size': self.input_size,
            'features': self.features,
            'is_image_set': True,
        }
        torch.save(res, path)

    def load_image_embedding(self, path):
        res = torch.load(path, self.device)
        for k, v in res.items():
            setattr(self, k, v)

You are my God ！