Frequency Domain Disentanglement for Arbitrary Neural Style Transfer
Authors: Dongyang Li, Hao Luo, Pichao Wang, Zhibin Wang, Shang Liu, Fan Wang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Qualitative and quantitative experiments verify that the proposed method can render better stylized results compared to the state-of-the-art methods. |
| Researcher Affiliation | Industry | Alibaba Group {yingtian.ldy, michuan.lh, pichao.wang, zhibin.waz, liushang.ls, fan.w}@alibaba-inc.com |
| Pseudocode | No | The paper describes its methods in text but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | Pytorch (Paszke et al. 2019) implementation of the Fre Mixer module is included in the supplementary material. |
| Open Datasets | Yes | We use MS-COCO (Lin et al. 2014) as the content image set and use Wiki Art (Phillips and Mackintosh 2011) as the style image set. |
| Dataset Splits | No | The paper states 'all test images were excluded from the training data' but does not explicitly specify train/validation/test dataset splits, proportions, or methods for creating them. |
| Hardware Specification | Yes | For the inference time, all the models are tested on a single NVIDIA Tesla V100-32G with batch size 1. |
| Software Dependencies | No | The paper mentions 'Pytorch (Paszke et al. 2019) implementation' but does not specify a concrete version number for PyTorch or other software dependencies. |
| Experiment Setup | Yes | During the training stage, all input images are first resized to 512 512 and then randomly cropped regions of size 256 256. Adam (Kingma and Ba 2014) with the learning rate of 0.0001 is used as the optimizer. We set the batch size to be 8 and train the proposed model with 160K iterations. |