Domain adaptive panoptic segmentation aims to mitigate data annotation challenge by leveraging off-the-shelf annotated data in one or multiple related source domains. However, existing studies employ two networks for instance segmentation and semantic segmentation separately which lead to a large amount of network parameters with complicated and computationally intensive training and inference processes. We design UniDAPS, a Unified Domain Adaptive Panoptic Segmentation network that is simple but capable of achieving domain adaptive instance segmentation and semantic segmentation simultaneously within a single network. UniDAPS introduces Hierarchical Mask Calibration (HMC) that rectifies the predicted pseudo masks, pseudo superpixels and pseudo pixels and performs network re-training via an online self-training process on the fly. It has three unique features: 1) it enables unified domain adaptive panoptic adaptation; 2) it mitigates false predictions and improves domain adaptive panoptic segmentation effectively; 3) it is end-to-end trainable with much less parameters and simpler training and inference pipeline. Extensive experiments over multiple public benchmarks show that UniDAPS achieves superior domain adaptive panoptic segmentation as compared with the state-of-the-art.

翻译：内地适应性全局光学截面旨在通过在一个或多个相关源域利用现成的附加说明数据来减轻数据注释化的挑战,然而,现有研究采用两个网络,例如分离和语义分离,分别导致大量具有复杂和计算密集的网络参数和推断过程;我们设计UniDAPS,一个简单但能够在单一网络内同时实现域适应性实例分离和语义分离的统一多面性光谱分割网;UniDAPS采用高端遮罩校准(HMC),对预测的假面罩、伪超级像素和伪像素进行校正,并通过在线自我培训过程对网络进行再培训;它有三个独特的特点:(1)它能够使统一的域适应性全局适应性适应适应性适应;(2)它减轻虚假预测,并有效地改进域适应性全局光谱分层;(3)它可以以低得多的参数和简化的培训和推断管道进行最终到终端的训练。