ETNet: Error Transition Network for Arbitrary Style Transfer

Authors: Chunjin Song, Zhijie Wu, Yang Zhou, Minglun Gong, Hui Huang

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Various qualitative and quantitative experiments show that the key concept of both progressive strategy and error-correction leads to better results. Code and models are available at https://github.com/zhijie W94/ETNet. In experiments, our model consistently outperforms the existing state-of-the-art models in qualitative and quantitative evaluations by a large margin. Ablation Study Our method has three key ingredients: iterative refinements, error measurement and the joint analysis between the error features and features of the intermediate stylization. Table 1 lists the quantitative metrics of the ablation study on above ingredients. Quantitative Comparison To quantitatively evaluate each method, we conduct a comparison regarding perceptual loss and report the results in the first two rows of Table 2.
Researcher Affiliation Academia Chunjin Song Shenzhen University songchunjin1990@gmail.com; Zhijie Wu Shenzhen University wzj.micker@gmail.com; Yang Zhou Shenzhen University zhouyangvcc@gmail.com; Minglun Gong University of Guelph minglun@uoguelph.ca; Hui Huang Shenzhen University hhzhiyan@gmail.com
Pseudocode No The paper provides architectural diagrams (Figure 2 and Figure 3) and describes the process using equations, but it does not include any structured pseudocode blocks or algorithms.
Open Source Code Yes Code and models are available at https://github.com/zhijie W94/ETNet. code will be made publicly available online.
Open Datasets No The paper mentions using content and style images and employs a VGG-based encoder, but it does not explicitly state the names of any public datasets used for training or provide access information for such datasets. It mentions a 'test set' for user studies, but not for training.
Dataset Splits No The paper refers to a 'test set' for evaluation but does not specify details about training, validation, or test splits (e.g., percentages, sample counts, or a specific splitting methodology).
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using a 'VGG-based encoder' but does not specify any software dependencies (e.g., deep learning frameworks, libraries, or programming languages) with their version numbers.
Experiment Setup Yes Thus the total loss at k-th level of a pyramid is computed as: Lk total = λk pc Lk pc + λk ps Lk ps + λk tv LT V , (9) where the λk pc and λk tv are always set to 1 and 10 6 while for k = 1, 2, 3, λk ps is assigned to 1, 5, 8 respectively to preserve semantic structures and gradually add style details to outputs.