SHUNIT: Style Harmonization for Unpaired Image-to-Image Translation
Authors: Seokbeom Song, Suhyeon Lee, Hongje Seong, Kyoungwon Min, Euntai Kim
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our method with extensive experiments and achieve state-of-the-art performance on the latest benchmark sets. The source code is available online: https://github.com/bluejangbaljang/SHUNIT. In this section, we present extensive experimental results and analysis. To demonstrate the superiority of our method, we compare our SHUNIT with state-of-the-art I2I translation methods. We evaluate our SHUNIT on three I2I translation scenarios: Cityscapes (Cordts et al. 2016) ACDC (Sakaridis, Dai, and Van Gool 2021) and INIT (Shen et al. 2019), and KITTI (Geiger et al. 2013) Cityscapes (Cordts et al. 2016). Table 1: Quantitative comparison on Cityscapes ACDC. We measure class-wise FID (lower is better) and m Io U (higher is better). Ablation Study In this section, we study the effectiveness of each component in our method. |
| Researcher Affiliation | Academia | 1Yonsei University, Seoul, Korea 2Korea Electronics Technology Institute, Seongnam, Korea {lgs5751, hyeon93, hjseong, etkim}@yonsei.ac.kr, minkw@keti.re.kr |
| Pseudocode | No | The paper describes the logic and components of the system through text and diagrams (Figure 2, 3, 4) but does not include a formal pseudocode block or algorithm. |
| Open Source Code | Yes | The source code is available online: https://github.com/bluejangbaljang/SHUNIT. |
| Open Datasets | Yes | We evaluate our SHUNIT on three I2I translation scenarios: Cityscapes (Cordts et al. 2016) ACDC (Sakaridis, Dai, and Van Gool 2021) and INIT (Shen et al. 2019), and KITTI (Geiger et al. 2013) Cityscapes (Cordts et al. 2016). Cityscapes (Cordts et al. 2016) is one of the most popular urban scene dataset. ACDC (Sakaridis, Dai, and Van Gool 2021) is the latest dataset with multiple adverse condition images and consists of four conditions of street scenes: snow, rain, fog, and night. INIT (Shen et al. 2019) is a public benchmark set for I2I translation. KITTI is a public benchmark set for object detection. |
| Dataset Splits | Yes | To train the networks, 2975, 400, 400, 400, and 400 images are used for clear, snow, rain, fog, and night conditions, respectively. Following (Shen et al. 2019), we split the 155K images into 85% for training and 15% for testing. We validate on Cityscapes (clear) ACDC (snow/rain) scenarios and use ACDC validation set for both class-wise FID and m Io U1. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU model, CPU type, memory) used for running the experiments. |
| Software Dependencies | No | The paper describes the use of various existing methods and components (e.g., Cycle GAN, MUNIT, PyTorch implied by typical ML research), but it does not specify any software dependencies with version numbers. |
| Experiment Setup | No | The paper mentions that 'The detailed explanations of those loss functions are given in the supplementary material' and 'The implementation details of our method are provided in the supplementary material.' It does not include specific hyperparameters (e.g., learning rate, batch size, number of epochs) or other system-level training settings in the main text. |