The paper introduces SID-Net, a novel single image dehazing network that addresses the limitations of existing methods by incorporating both adversarial and contrastive learning. Image dehazing is a crucial low-level vision task that aims to restore clear images from hazy ones, which are often degraded by atmospheric conditions. Most existing methods focus on clean images and fail to utilize negative information from hazy images, leading to sub-optimal performance. SID-Net consists of three core branches: Image Dehazing Branch (IDB), Adversarial Guidance Branch (AGB), and Contrastive Enhancement Branch (CEB). IDB uses an encoder-decoder framework with an Attentive Recurrent Module (ARM) and Attention Fusion Operation (AFO) to transform hazy images into clean ones. AGB leverages positive information from clean ground truth images through adversarial learning to guide the restoration process. CEB employs contrastive learning to exploit negative information from hazy images, enhancing the dehazing performance. Experiments on synthetic and real-world datasets show that SID-Net achieves comparable results to state-of-the-art algorithms. The code for SID-Net is available on GitHub.The paper introduces SID-Net, a novel single image dehazing network that addresses the limitations of existing methods by incorporating both adversarial and contrastive learning. Image dehazing is a crucial low-level vision task that aims to restore clear images from hazy ones, which are often degraded by atmospheric conditions. Most existing methods focus on clean images and fail to utilize negative information from hazy images, leading to sub-optimal performance. SID-Net consists of three core branches: Image Dehazing Branch (IDB), Adversarial Guidance Branch (AGB), and Contrastive Enhancement Branch (CEB). IDB uses an encoder-decoder framework with an Attentive Recurrent Module (ARM) and Attention Fusion Operation (AFO) to transform hazy images into clean ones. AGB leverages positive information from clean ground truth images through adversarial learning to guide the restoration process. CEB employs contrastive learning to exploit negative information from hazy images, enhancing the dehazing performance. Experiments on synthetic and real-world datasets show that SID-Net achieves comparable results to state-of-the-art algorithms. The code for SID-Net is available on GitHub.