Beyond Spatial Domain: Cross-domain Promoted Fourier Convolution Helps Single Image Dehazing
Authors: Xiaozhe Zhang, Haidong Ding, Fengying Xie, Linpeng Pan, Yue Zi, Ke Wang, Haopeng Zhang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Comprehensive experiments on multiple datasets demonstrate that JSFC-Net is significantly superior to SOTA dehazing methods. Comprehensive quantitative and qualitative experimental results demonstrate the superiority of JSFC-Net over SOTA methods. Extensive experiments demonstrate that JSFC-Net significantly outperform existing SOTA methods in various benchmarks, particularly in real-world dehazing. |
| Researcher Affiliation | Academia | Xiaozhe Zhang1,2, Haidong Ding1,2, Fengying Xie1,2*, Linpeng Pan1,2, Yue Zi3, Ke Wang1,2, Haopeng Zhang1,2 1School of Astronautics, Beihang University 2Tianmushan Laboratory, Beihang University 3School of Electrical and Information Engineering, Changsha University of Science and Technology EMAIL, EMAIL |
| Pseudocode | No | The paper describes the architecture and components of JSFC-Net using text and diagrams (Figure 2), but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code for the described methodology, nor does it include links to code repositories or mention code in supplementary materials. |
| Open Datasets | Yes | For a comprehensive comparison, we evaluate the proposed JSFC-Net on synthetic haze datasets (i.e., RESIDE (Li et al. 2018)), generated haze datasets (i.e., NHHAZE (Ancuti, Ancuti, and Timofte 2020) and O-Haze (Ancuti et al. 2018) datasets) and real-world hazy images. |
| Dataset Splits | Yes | In RESIDE (Li et al. 2018), two subsets, the Indoor Training Set (ITS) and the Outdoor Training Set (OTS), are selected for training, consisting of 13,990 pairs and 313,950 pairs of images, respectively. Models are evaluated on the Synthetic Objective Testing Set (SOTS) subset. The NH-HAZE (Ancuti, Ancuti, and Timofte 2020) and O-Haze (Ancuti et al. 2018) datasets contain 55 and 45 images, respectively. We select 5 images from each dataset as test sets and the remaining images as training sets. |
| Hardware Specification | Yes | The JSFC-Net is implemented using the Py Torch framework on an Intel Gold 6252 CPU and NVIDIA A100 GPUs. |
| Software Dependencies | No | The paper mentions "Py Torch framework" but does not specify a version number for it or any other key software libraries or dependencies. |
| Experiment Setup | Yes | We use the Adam (Kingma and Ba 2014) optimizer with default parameters (β1 = 0.9, β2 = 0.99) and a cosine annealing strategy (Loshchilov and Hutter 2016) to train JSFC-Net. The initial learning rate is set to 1 10 4 and gradually decreases to 5 10 6. |