During image editing, existing deep generative models tend to re-synthesize the entire output from scratch, including the unedited regions. This leads to a significant waste of computation, especially for minor editing operations. In this work, we present Spatially Sparse Inference (SSI), a general-purpose technique that selectively performs computation for edited regions and accelerates various generative models, including both conditional GANs and diffusion models. Our key observation is that users tend to make gradual changes to the input image.

During image editing, existing deep generative models tend to re-synthesize the entire output from scratch, including the unedited regions. This leads to a significant waste of computation, especially for minor editing operations. In this work, we present Spatially Sparse Inference (SSI), a general-purpose technique that selectively performs computation for edited regions and accelerates various generative models, including both conditional GANs and diffusion models.

During image editing, existing deep generative models tend to re-synthesize the entire output from scratch, including the unedited regions. This leads to a significant waste of computation, especially for minor editing operations. In this work, we present Spatially Sparse Inference (SSI), a general-purpose technique that selectively performs computation for edited regions and accelerates various generative models, including both conditional GANs and diffusion models.

During image editing, existing deep generative models tend to re-synthesize the entire output from scratch, including the unedited regions. This leads to a significant waste of computation, especially for minor editing operations. In this work, we present Spatially Sparse Inference (SSI), a general-purpose technique that selectively performs computation for edited regions and accelerates various generative models, including both conditional GANs and diffusion models. Our key observation is that users tend to make gradual changes to the input image. Given an edited image, we sparsely apply the convolutional filters to the edited regions while reusing the cached features for the unedited regions.

Abstract: Due to the difficulty of collecting electrocardiogram (ECG) data during emergency situations, ECG data generation is an efficient solution for dealing with highly imbalanced ECG training datasets. However, due to the complex dynamics of ECG signals, the synthesis of such signals is a challenging task. In this paper, we present a novel approach for ECG signal generation based on Generative Adversarial Networks (GANs). Our approach combines GANs with statistical ECG data modeling to leverage prior knowledge about ECG dynamics in the generation process. To validate the proposed approach, we present experiments using ECG signals from the MIT-BIH arrhythmia database.