This paper proposes an adversarial learning method for domain adaptation in semantic segmentation. The method adapts the output space of semantic segmentations, which contain spatial similarities between source and target domains. A multi-level adversarial network is constructed to perform output space domain adaptation at different feature levels. Extensive experiments on synthetic-to-real and cross-city scenarios show that the proposed method performs favorably against state-of-the-art methods in terms of accuracy and visual quality. The method adapts features through back-propagation from output labels, and a multi-level adversarial learning strategy is developed to adapt features at different levels of the segmentation model. The method is evaluated on various benchmark datasets, including GTA5, SYNTHIA, and Cross-City, showing improved performance over single-level adaptation. The contributions include proposing a domain adaptation method for pixel-level semantic segmentation via adversarial learning, demonstrating that adaptation in the output space can effectively align scene layout and local context between source and target images, and developing a multi-level adversarial learning scheme to adapt features at different levels of the segmentation model, leading to improved performance.This paper proposes an adversarial learning method for domain adaptation in semantic segmentation. The method adapts the output space of semantic segmentations, which contain spatial similarities between source and target domains. A multi-level adversarial network is constructed to perform output space domain adaptation at different feature levels. Extensive experiments on synthetic-to-real and cross-city scenarios show that the proposed method performs favorably against state-of-the-art methods in terms of accuracy and visual quality. The method adapts features through back-propagation from output labels, and a multi-level adversarial learning strategy is developed to adapt features at different levels of the segmentation model. The method is evaluated on various benchmark datasets, including GTA5, SYNTHIA, and Cross-City, showing improved performance over single-level adaptation. The contributions include proposing a domain adaptation method for pixel-level semantic segmentation via adversarial learning, demonstrating that adaptation in the output space can effectively align scene layout and local context between source and target images, and developing a multi-level adversarial learning scheme to adapt features at different levels of the segmentation model, leading to improved performance.