This paper proposes a semi-supervised nighttime image dehazing baseline, SFSNiD, which addresses the challenges of nighttime haze, glow, and noise with localized, coupled, and frequency inconsistent characteristics. The proposed method integrates spatial and frequency domain information through a spatial-frequency information interaction (SFII) module, which dynamically filters and aggregates frequency spectrum features. A retraining strategy and local window-based brightness loss are designed to suppress haze and glow while achieving realistic brightness. The method is validated on public benchmarks, demonstrating its effectiveness and superiority over state-of-the-art methods. The SFII module dynamically processes frequency spectrum features, while the retraining strategy utilizes pseudo labels to enhance performance. The method also incorporates a realistic brightness constraint, ensuring the dehazed images maintain realistic brightness levels. Experiments show that the proposed method achieves better performance in terms of both quantitative and visual results compared to existing methods. The method is implemented using PyTorch and trained on a single NVIDIA RTX 4090 platform. The results show that the proposed method effectively handles the challenges of nighttime dehazing, achieving better performance in terms of both quantitative and visual results compared to existing methods.This paper proposes a semi-supervised nighttime image dehazing baseline, SFSNiD, which addresses the challenges of nighttime haze, glow, and noise with localized, coupled, and frequency inconsistent characteristics. The proposed method integrates spatial and frequency domain information through a spatial-frequency information interaction (SFII) module, which dynamically filters and aggregates frequency spectrum features. A retraining strategy and local window-based brightness loss are designed to suppress haze and glow while achieving realistic brightness. The method is validated on public benchmarks, demonstrating its effectiveness and superiority over state-of-the-art methods. The SFII module dynamically processes frequency spectrum features, while the retraining strategy utilizes pseudo labels to enhance performance. The method also incorporates a realistic brightness constraint, ensuring the dehazed images maintain realistic brightness levels. Experiments show that the proposed method achieves better performance in terms of both quantitative and visual results compared to existing methods. The method is implemented using PyTorch and trained on a single NVIDIA RTX 4090 platform. The results show that the proposed method effectively handles the challenges of nighttime dehazing, achieving better performance in terms of both quantitative and visual results compared to existing methods.