Deep Fourier Up-Sampling

Authors: man zhou, Hu Yu, Jie Huang, Feng Zhao, Jinwei Gu, Chen Change Loy, Deyu Meng, Chongyi Li

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments across multiple computer vision tasks, including object detection, image segmentation, image de-raining, image dehazing, and guided image superresolution, demonstrate the consistent performance gains obtained by introducing our Fourier Up.
Researcher Affiliation Collaboration Man Zhou1,2 , Hu Yu2 , Jie Huang2, Feng Zhao2, Jinwei Gu6, Chen Change Loy3, Deyu Meng4,5, Chongyi Li3 1Hefei Institute of Physical Science, Chinese Academy of Sciences, China 2University of Science and Technology of China, China 3S-Lab, Nanyang Technological University, Singapore 4Xi an Jiaotong University, China 5Pazhou Laboratory (Huangpu), China 6Sense Brain Technology (Sense Time Research USA), USA
Pseudocode Yes Figure 2: Pseudo-code of the two variants of the proposed deep Fourier up-sampling. The left is the periodic padding variant while the right is the area interpolation-cropping variant.
Open Source Code Yes Code is available at https://manman1995.github.io/.
Open Datasets Yes Following [12], the PASCAL VOC 2007 and 2012 training sets [37] are used as training data.
Dataset Splits No The PASCAL VOC 2007 testing set is used for evaluations as the ground truth annotations of VOC 2012 testing set are not publicly available.
Hardware Specification Yes Ascend AI Processor used for this research.
Software Dependencies No Mind Spore, CANN, the JKW Research Funds under Grant 20-163-14-LZ-001-004-01 and Ascend AI Processor used for this research.
Experiment Setup No For a fair comparison, we use the same number of trainable parameters as 2)/3).