Spherical Pseudo-Cylindrical Representation for Omnidirectional Image Super-resolution

Authors: Qing Cai, Mu Li, Dongwei Ren, Jun Lyu, Haiyong Zheng, Junyu Dong, Yee-Hong Yang

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

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results on public datasets demonstrate the effectiveness of the proposed method as well as the consistently superior performance of our method over most state-of-the-art methods both quantitatively and qualitatively.
Researcher Affiliation Academia 1 Faculty of Computer Science and Technology, Ocean University of China 2 School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen 3 School of Computer Science and Technology, Harbin Institute of Technology 4 School of Nursing, The Hong Kong Polytechnic University 5 Department of Computing Science, University of Alberta
Pseudocode No The paper describes procedural steps in text but does not include a formal pseudocode or algorithm block.
Open Source Code No The paper mentions retraining comparison methods using their open-source codes, but does not provide an explicit statement or link for the open-source code of the proposed method.
Open Datasets Yes Following previous methods (Yoon et al. 2022), we also choose ODI-SR (Deng et al. 2021) as our training dataset, which contains 1200 training images, 100 validation images, and 100 testing images. We use the ODI-SR and SUN 360 Panorama (Xiao et al. 2012) as our test datasets.
Dataset Splits Yes Following previous methods (Yoon et al. 2022), we also choose ODI-SR (Deng et al. 2021) as our training dataset, which contains 1200 training images, 100 validation images, and 100 testing images.
Hardware Specification Yes The whole process is implemented in the PyTorch platform with 4 RTX3090 GPUs, each with 24GB of memory (Please see the supplementary for more details).
Software Dependencies No The paper mentions 'PyTorch platform' but does not specify its version number or any other software dependencies with specific versions.
Experiment Setup Yes Following previous works (Deng et al. 2021; Yoon et al. 2022), we train our model for the scales of 8 and 16, and all degraded datasets are obtained using bicubic interpolation. To avoid boundary artifacts between neighboring tiles, following previous work (Deng et al. 2021), an extra Ht/8 is added for neighboring tiles, where Ht denotes the height of each tile. The proposed model is trained by the ADAM optimizer (Kingma and Ba 2014) with a fixed initial learning rate of 10-4.