2ffbfa3a6bd3b8de8e21a762489346054dcd9ccc,nussl/separation/deep/deep_mask_estimation.py,DeepMaskEstimation,extract_features,#DeepMaskEstimation#,53

Before Change

        input_data = self._preprocess()
        with torch.no_grad():
            features = self.model(input_data)
            if "estimates" not in features:
                raise ValueError("This model is not a mask estimation model!")
            features = (
                (features["estimates"] + 1e-6) / 
                (input_data["magnitude_spectrogram"].unsqueeze(-1) + 1e-6)
            ).squeeze(0)
        features = features.permute(3, 1, 0, 2)
        features = features.data.cpu().numpy()
        return features

    def run(self, features=None):
        

After Change

            embedding = output["embedding"]
            // swap back batch and sample dims
            if self.metadata["num_channels"] == 1:
                embedding = embedding.transpose(0, -2)
            embedding = embedding.squeeze(0).transpose(0, 1)
        return embedding.cpu().data.numpy()

Instances