[Re] Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization

Mert Erkol, Furkan Kınlı, Barış Özcan, Furkan Kıraç

Keywords: feature distribution matching, style transfer, domain generalization

TL;DR: In this reproducibility study, we present our results and experience during replicating the paper, titled Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization.

Abstract:

Reproducibility Summary:

In this reproducibility study, we present our results and experience during replicating the paper, titled Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization [1]. In real‐world scenarios, the feature distributions are mostly much more complicated than Gaussian, so only mean and standard deviation may not be fully representative to match them. This paper introduces a novel strategy to exactly match the histograms of image features via the Sort‐Matching algorithm in a computationally feasible way. We were able to reproduce most of the results presented in the original paper both qualitatively and quantitatively.

Scope of Reproducibility — In the scope of this study, we aim to reproduce all the qualitative and quantitative results on two tasks, namely Arbitrary Style Transfer (AST) and Domain Generalization (DG). Moreover, we investigate the capability of forming better style representations by EFDM in another recent study [2].

Methodology — We have conducted all experiments in the original work by using the official repository, which is implemented by PyTorch [3]. For additional experiments, we have implemented the modular version of EFDM as a layer to replace it with the normalization modules. We have used 2 NVIDIA RTX 2080Ti GPUs for both training and testing, and it took roughly 1 day to complete a single training.

Results — We have reproduced the experiments done on two selected tasks, and compared their results with the reported results. Although our experimental results are not identical to the reported ones, we can validate the claims made by the original study according to these results.

What was easy — The paper is well‐written and easy to follow. The original repository is well‐organized to run all tests with the data presented in the paper.

What was difficult — The requirements in the repository were not updated, and we had to manage different versions of Python packages to be able to conduct the experiments.

Communication with original authors — We were in contact with the authors, and asked for the original results as JPEG files to prepare the figures in this report.

Paper Url: https://openaccess.thecvf.com/content/CVPR2022/papers/Zhang_Exact_Feature_Distribution_Matching_for_Arbitrary_Style_Transfer_and_Domain_CVPR_2022_paper.pdf

Paper Venue: CVPR 2022

Confirmation: The report pdf is generated from the provided camera ready Google Colab script, The report metadata is verified from the camera ready Google Colab script, The report contains correct author information., The report contains link to code and SWH metadata., The report follows the ReScience latex style guides as in the Reproducibility Report Template (https://paperswithcode.com/rc2022/registration)., The report contains the Reproducibility Summary in the first page., The latex .zip file is verified from the camera ready Google Colab script

Journal: ReScience Volume 9 Issue 2 Article 2

Doi: https://www.doi.org/10.5281/zenodo.8173652

Code: https://archive.softwareheritage.org/swh:1:dir:b76a5bf3f3f540d17ef0f4a22ecc0b4e2c27d680