/Neural_Color_Transfer

Implementation of Neural Color Transfer between Images by PyTorch.

Primary LanguageJupyter Notebook

Neural_Color_Transfer_PyTorch_Implementation

In progress

We are trying to implement Neural Color Transfer between Images by Mingming He et al in PyTorch. They revised their paper to version 2 which is Progressive Color Transfer with Dense Semantic Correspondences.

Paper

Progressive Color Transfer with Dense Semantic Correspondences

Abstract

We propose a new algorithm for color transfer between images that have perceptually similar semantic structures. We aim to achieve a more accurate color transfer that leverages semantically-meaningful dense correspondence between images. To accomplish this, our algorithm uses neural representations for matching. Additionally, the color transfer should be spatially variant and globally coherent. Therefore, our algorithm optimizes a local linear model for color transfer satisfying both local and global constraints. Our proposed approach jointly optimizes matching and color transfer, adopting a coarse-to-fine strategy. The proposed method can be successfully extended from one-to-one to one-to-many color transfer. The latter further addresses the problem of mismatching elements of the input image. We validate our proposed method by testing it on a large variety of image content.

Process

Implemented on Single Reference Neural Color Transfer And changed WLS-based filter to Deep Guided Filter.

Pipeline

K-006

Results

Input Image

Input

Style image

Style

And the result from those images are here. The images are from layer 5 to 1 below.

L=5

img5S

L=4

img4S

L=3

img3S

L=2

img2S

L=1

img1S

TODO

  • Need to modularize our notebook implementation.
  • Our guidance and result image is different from original paper so need to fix some codes.
  • The performance speed is way too slow than original paper.
  • If the single reference is working well, need to work on multi-reference.
  • Refine MarkDown