Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Pan-LUT: Efficient Pan-sharpening via Learnable Look-Up Tables
Authors: Zhongnan Cai, Yingying Wang, Hui Zheng, Panwang Pan, Zixu Lin, Ge Meng, Chenxin Li, Chunming He, Jiaxin Xie, Yunlong Lin, Junbin Lu, Yue Huang, Xinghao Ding
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Experiments Remote sensing datasets from three satellites are used in our experiments, including World View II (WV2), Gao Fen2 (GF2) and World View-III (WV3). ... 4.3 Comparison with Other Methods Evaluation on Reduced-resolution Scene. The quantitative results across three datasets are presented in Table 1, with the best results highlighted in red. 4.4 Ablation Study Size of Look-Up Tables. As shown in Figure 5, changing the LUT size does not lead to a significant drop in performance. |
| Researcher Affiliation | Collaboration | 1Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, Xiamen, Fujian, China 2Byte Dance 3The Chinese University of Hong Kong 4Duke University 5University of Washington |
| Pseudocode | No | The paper uses mathematical formulas and descriptions of the proposed LUTs but does not include a distinct section labeled 'Pseudocode' or 'Algorithm' with structured steps. |
| Open Source Code | Yes | The source code is available at https://github.com/CZhongnan/Pan-LUT. |
| Open Datasets | Yes | Remote sensing datasets from three satellites are used in our experiments, including World View II (WV2), Gao Fen2 (GF2) and World View-III (WV3). Due to the absence of high-resolution multispectral ground truth images in these datasets, we generate the training set using the Wald protocol tool [37]. |
| Dataset Splits | Yes | Specifically, given the original MS image and its corresponding high-resolution PAN image, they are downsampled by a factor of r to obtain image pairs of MS and PAN, with r set to 4. During training, the original high-resolution MS image is treated as the ground truth, while the MS and PAN images serve as the input image pairs. To assess the performance and generalization capability of our method on full-resolution scenes under real-world conditions, we first trained Pan-LUT on the reduced-resolution World View-II data and then tested it on unseen full-resolution World View-II satellite datasets. The real-world dataset consists of 200 newly collected samples from the World View II satellite for evaluation. |
| Hardware Specification | Yes | Our method makes it possible to process 15K 15K remote sensing images on a 24GB GPU. ... Our model contains fewer than 700K parameters and processes a 9K 9K image in under 1 ms using one RTX 2080 Ti GPU... In the CPU inference time experiments, all methods were conducted on a workstation equipped with an Intel(R) Xeon(R) Gold 6226R CPU. |
| Software Dependencies | No | The PyTorch framework is implemented in our experiment. |
| Experiment Setup | Yes | During the training phase, we employ an ADAM optimizer with β1 = 0.9 and β2 = 0.999, to update the network parameters for 1000 epochs with a batch size of 1. The learning rate is initialized with 5 10 4. In parallel, a Step LR learning rate adjustment strategy is employed to reduce the learning rate by half after every 200 iterations. The sizes of PGLUT, SDLUT and AOLUT are set to 9, 9 and 9, respectively. In our experiments, we empirically set λs = 0.0001 and λm = 10. |