modulated_conv2d

Modulated Convolutional Neural Networks (Modulated Convolution or ModConv) is a type of convolutional layer in deep learning that uses a learned modulation parameter to modulate the activation of the convolutional filters.

In traditional convolutional neural networks, the filters are fixed and applied uniformly across the entire input image. However, in modulated convolution, the filters are dynamically modulated based on the input image features. This allows the network to adapt the filters to the local features of the input image, leading to improved performance.

Modulated convolution is commonly used in image generation tasks, such as image super-resolution, style transfer, and image synthesis. It was first introduced in the paper "Modulated Convolutional Networks" by Dario R. Coelho and Tiago A. Almeida in 2019.