Who Looks like Me: Semantic Routed Image Harmonization

Authors: Jinsheng Sun, Chao Yao, Xiaokun Wang, Yu Guo, Yalan Zhang, Xiaojuan Ban

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our method has achieved excellent experimental results on existing datasets and our model outperforms the stateof-the-art by a margin of 0.45 d B on i Harmony4 dataset.
Researcher Affiliation Academia 1Beijing Advanced Innovation Center for Materials Genome Engineering, School of Intelligence Science and Technology, University of Science and Technology Beijing, Beijing 100083, China 2School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China 3Key Laboratory of Intelligent Bionic Unmanned Systems, Ministry of Education, University of Science and Technology Beijing, Beijing 100083, China 4Institute of Materials Intelligent Technology, Liaoning Academy of Materials, Shenyang 110004, China
Pseudocode No The paper describes the model architecture and processes in detail but does not include any explicit pseudocode blocks or algorithms labeled as such.
Open Source Code Yes Our code is available in github.
Open Datasets Yes Our experiments use the i Harmony4 dataset, a publicly available synthesized dataset referenced by Cong et al. [Cong et al., 2020], which includes four sub-datasets: HCOCO, HAdobe5k, HFlickr, and Hday2night.
Dataset Splits Yes We employed the same processing method as HDNet [Chen et al., 2022] for the dataset. Additionally, to validation the performance of our methods in real-world scenarios, we employed 100 real-world images from CDTNet [Cong et al., 2022], which are processed in the format of the i Harmony4 dataset.
Hardware Specification Yes We use Py Torch to implement our models with NVIDIA Ge Force RTX 4090.
Software Dependencies No The paper mentions 'Py Torch' as the implementation framework but does not specify a version number for it or any other software dependencies, which is required for reproducibility.
Experiment Setup Yes Our model is trained by Adam W optimizer with β1 = 0.9, β2 = 0.999, and weight decay 1e 4. We train the model for 200 epochs with input images resized to 256 256 and batch size set to 8. The initial learning rate is set to 3e 4 and gradually reduced to 1e 6 with the cosine annealing [Loshchilov and Hutter, 2017].