Unsupervised Domain Adaptation (UDA) mitigates the domain shift between the source and target domains. Some conventional methods are based on guiding semantic segmentation across domains using depth information. Nevertheless, these methods are constrained and inadequate for the use of depth knowledge. In this work, we propose a Depth-aware Adaptation Framework (DAF) and an Intra-domain Adaptation (IDA) strategy to improve semantic segmentation in the context of UDA. We design a novel depth estimation network based on the channel-wise attention mechanism to provide additional depth information. In our approach, DAF aims to adapt domains by exploiting the inherent correlations of semantic and depth information, where we construct a depth-aware space for the alignment between source and target domains. Furthermore, IDA bridges the gap within the target domain through a proposed depth-aware ranking strategy. This strategy is based on dividing the target domain into a sub-source domain and a sub-target domain. The distribution discrepancies between these domains are then aligned. Extensive experiments on SYNTHIA→Cityscapes and SYNTHIA→Mapillary cross-domain tasks confirm that our method overperforms the conventional depth guidance methods with mean IOU values of 46.7% and 73.3% on the above two tasks, respectively.
Estimating the 6-D poses of objects from RGB-D images holds great potential for several applications. However, given that the 6-D pose estimation accuracy is significantly affected by occlusion and noise between the objects in an image, this paper proposes a novel 6-D pose estimation method based on Efficient feature extraction and Point-pair feature matching. Specifically, we develop the Efficient channel attention Convolutional Neural Network (ECNN) and SO(3)-Encoder modules to extract 2-D features from the RGB image and SO(3)-equivariant features from the depth image, respectively. These features are fused in the DenseFusion module to obtain 3-D features in the camera space. Meanwhile, we exploit CAD model priors to obtain 3-D features in the model space through the model feature encoder, and then we globally regress the 3-D features in the camera and model space. According to these features, we generate oriented point clouds in each space, and then conduct point-pair feature matching to obtain pose information. Finally, we perform direct pose regression on the 3-D features in the camera and model space, and then resulting point-pair feature matching pose information is combined with the direct point-wise pose regression information to enhance pose prediction accuracy. Experimental results on three widely used benchmarking datasets demonstrate that our method achieves state-of-the-art performance, particularly for severe occluded scenes. Our code is available at https://github.com/Karthus0930/EPM-Net