Focus Stacking with High Fidelity and Superior Visual Effects
Authors: Bo Liu, Bin Hu, Xiuli Bi, Weisheng Li, Bin Xiao
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Based on extensive experiments on simulated dataset, existing realistic dataset and our proposed Beta Fusion dataset, the results show that our proposed method can generate high-quality all-in-focus images by achieving two goals simultaneously, especially can successfully solve the TR problem and eliminate the visual effect degradation of synthesized images caused by the TR problem. |
| Researcher Affiliation | Academia | Department of Computer Science and Technology, Chongqing University of Posts and Telecommunications, Chongqing 400065, China {boliu, bixl, liws, xiaobin}@cqupt.edu.cn, s210201036@stu.cqupt.edu.cn |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks (i.e., a figure, block, or section labeled "Pseudocode" or "Algorithm"). |
| Open Source Code | No | The paper does not include an unambiguous statement that the authors are releasing the source code for the work described, nor does it provide a direct link to a code repository. |
| Open Datasets | Yes | Due to the scarcity of focus stacking datasets, we used a simulated dataset for training and testing, where the foregrounds were taken from the AM2K (Li et al. 2022) dataset, and the backgrounds were taken from the DIV2K (Agustsson and Timofte 2017) dataset. Additionally, we used two real-world focus stacking datasets in experiments: Lytro Dataset (Nejati, Samavi, and Shirani 2015) and our Beta Fusion dataset, both contain 20 image pairs. |
| Dataset Splits | No | The paper specifies a split for training (19.8k) and testing (1.8k) for the simulated dataset, but does not explicitly mention a separate validation split or its size. |
| Hardware Specification | Yes | The training was with Nvidia RTX A6000 with 48GB video memory. |
| Software Dependencies | No | The paper mentions the use of an "Adam W optimizer" but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, or specific library versions). |
| Experiment Setup | Yes | The batch size is 32, and the initial learning rate of the Adam W optimizer is set to 1e-4 and will be halved every 25 epochs. The whole training process lasts for about 400 epochs. Parameter λ1 and λ2 in Eq.11 are set to 2 and 8, respectively. ϵ in Eq.13 is set to 1e-10. |