Coarse-to-Fine Image Inpainting via Region-wise Convolutions and Non-Local Correlation
Authors: Yuqing Ma, Xianglong Liu, Shihao Bai, Lei Wang, Dailan He, Aishan Liu
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three widely-used datasets for image inpainting tasks have been conducted, and both qualitative and quantitative experimental results demonstrate that the proposed model significantly outperforms the state-of-the-art approaches, especially for the large irregular missing regions. |
| Researcher Affiliation | Academia | 1State Key Lab of Software Development Environment, Beihang University, China 2Beijing Advanced Innovation Center for Big Data-Based Precision Medicine, Beihang University, Beijing, China |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper provides a link to supplementary material which may contain results, and mentions borrowing an implementation for a baseline method (PConv) from GitHub. However, it does not explicitly state that *their* source code for the proposed methodology is made publicly available. |
| Open Datasets | Yes | We employ the widely-used datasets in prior studies, including Celeb A-HQ [Karras et al., 2017], Places2 [Zhou et al., 2018], and Paris Street View [Doersch et al., 2012]. |
| Dataset Splits | Yes | Celeb A-HQ contains 30k high-resolution face images, and we adopt the same partition as [Yu et al., 2018b] did. The Places2 dataset includes 8,097,967 training images with diverse scenes. The Paris Street View contains 14,900 training images and 100 test images. For both datasets, we adopt the original train, test, and validate splits. |
| Hardware Specification | No | The paper does not provide specific hardware details (like exact GPU/CPU models, memory, or specific cloud instances) used for running its experiments. |
| Software Dependencies | No | The paper mentions using the "Adam optimizer" and a "pre-trained VGG network" but does not provide specific version numbers for these or any other software libraries or frameworks used. |
| Experiment Setup | Yes | Input images are resized to 256 256, and the proportion of irregular missing regions varies from 0 to 40% in the training process. We empirically choose the hyper-parameters λ1 = 10−5, λ2 = 10−3, and the initial learning rate 10−4. Using the Adam optimizer, on Celeb A-HQ and Paris Street View we train the model with a batch size of 8 for 20 epochs, and on Places2 we train it with a batch size of 48. |