Rubik's Cube: High-Order Channel Interactions with a Hierarchical Receptive Field

Authors: Naishan Zheng, man zhou, Chong Zhou, Chen Change Loy

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted experiments across various low-level vision tasks, including image denoising, low-light image enhancement, guided image super-resolution, and image de-blurring. The results consistently demonstrate that our Rubik s cube operator enhances performance across all tasks.
Researcher Affiliation Academia Naishan Zheng Man Zhou* Chong Zhou Chen Change Loy S-Lab, Nanyang Technological University naishan.zheng,man.zhou,chong033,ccloy@ntu.edu.sg
Pseudocode Yes The details of the designed operator and its pseudo-code are summarized in Figure 2 The proposed Rubik s cube convolution achieves high-order channel interaction and information aggregation by element-wise multiplication and convolution layers with 1 1 kernel, which is simple to implement.
Open Source Code Yes Code is publicly available at https://github.com/zheng980629/Rubik Cube.
Open Datasets Yes We conducted experiments on two widely used low-light image enhancement datasets: LOL [43] and Huawei [44]. The LOL dataset consists of 500 paired low-/normal images, and we split 485 for training and 15 for testing as the official selection. For the Huawei dataset, we randomly selected 2,200 images for training and kept the remaining 280 images for testing purposes.
Dataset Splits No The LOL dataset consists of 500 paired low-/normal images, and we split 485 for training and 15 for testing as the official selection. For the Huawei dataset, we randomly selected 2,200 images for training and kept the remaining 280 images for testing purposes.
Hardware Specification No No specific hardware details (like GPU models, CPU types, or memory specifications) used for running experiments are mentioned in the paper.
Software Dependencies No The paper refers to 'PyTorch-style code' in Figure 2 but does not specify PyTorch's version or any other software dependencies with their version numbers.
Experiment Setup Yes In this work, the default shift of pixel is set to 1, and its robustness is validated in Sec. 4.4. Therefore, in all the experiments presented in this paper, we set p to 1 as the default value for the number of shifted pixels. We have set the default ratio r to 1/2 for all experiments conducted in this study.