taki0112 / AdaConv-Tensorflow

Simple Tensorflow implementation of "Adaptive Convolutions for Structure-Aware Style Transfer" (CVPR 2021)

AdaConv — Simple TensorFlow Implementation [Paper]

: Adaptive Convolutions for Structure-Aware Style Transfer (CVPR 2021)

Note

This repository does not implement all codes, but only implements the core modules of the paper.

Requirements

• Tensorflow == 2.5.0

Usage

feats = tf.random.normal(shape=[5, 64, 64, 256])
style_w = tf.random.normal(shape=[5, 512])

kp = KernelPredict(in_channels=feats.shape[-1], group_div=1)
adac = AdaConv(channels=1024, group_div=1)

w_spatial, w_pointwise, bias = kp(style_w)
x = adac([feats, w_spatial, w_pointwise, bias]) # [5, 64, 64, 1024]

Reference

• https://github.com/RElbers/ada-conv-pytorch

Author

Junho Kim