Deep Unfolded Network with Intrinsic Supervision for Pan-Sharpening
Authors: Hebaixu Wang, Meiqi Gong, Xiaoguang Mei, Hao Zhang, Jiayi Ma
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the advantages of our method compared to state-of-the-arts, showcasing its remarkable generalization capability to real-world scenes. |
| Researcher Affiliation | Academia | Electronic Information School, Wuhan University, Wuhan 430072, China {wanghebaixu, meixiaoguang, zhpersonalbox, jyma2010}@gmail.com, meiqigong@whu.edu.cn |
| Pseudocode | No | Following the framework of half-quadratic splitting (HQS) (Sun et al. 2020), two auxiliary variables U and V are introduced to reformulate Eq. (2): arg min H,U,V 1 2 L DBH 2 2 + η1 2 U H 2 2 + η2 2 V H 2 2 + λ2 2 Ω2(P, V ), (3) where η1, η2, λ1 and λ2 are penalty parameters. To achieve the unrolling inference, Eq. (3) can be divided into the following three sub-problems and solved alternatively: U (k) = arg min U η1 U H(k) 2 2 + η2 Tp P Th U 2 2, (4) V (k) = arg min V λ1 V H(k) 2 2 + λ2 DBV IP 2 2, (5) H(k+1) = arg min H 1 2 L DBH 2 2 + η1 U (k) H 2 2 V (k) H 2 2, (6) |
| Open Source Code | Yes | Our code is publicly available at https://github.com/Baixuzx7/DISPNet. |
| Open Datasets | Yes | Extensive experiments are conducted over three satellite datasets, namely Gao Fen-2, Quickbird and World View-II. ... In the training stage, the reduced image pairs are treated as inputs, while H is regarded as a reference. |
| Dataset Splits | No | In the training stage, the reduced image pairs are treated as inputs, while H is regarded as a reference. |
| Hardware Specification | Yes | All the experiments are conducted on a desktop with 2.6GHz AMD EPYC 7H12, NVIDIA Ge Force RTX 3090. |
| Software Dependencies | No | The implementation is based on the Pytorch framework. |
| Experiment Setup | Yes | For optimization, the learning rate is set to 1 10 4. The Adam optimizer is employed with to update the network parameters for 600 epochs with the batch size of 16. The number of unfolding stages is K = 4, other coefficients are α=0.1, β =1, γ =0.01, ρ=0.1 and λ= 10. |