In recent studies, self-supervised learning methods have been explored for monocular depth estimation. They minimize the reconstruction loss of images instead of depth information as a supervised signal. However, existing methods usually assume that the corresponding points in different views should have the same color, which leads to unreliable unsupervised signals and ultimately damages the reconstruction loss during the training. Meanwhile, in the low texture region, it is unable to predict the disparity value of pixels correctly because of the small number of extracted features. To solve the above issues, we propose a network—PDANet—that integrates perceptual consistency and data augmentation consistency, which are more reliable unsupervised signals, into a regular unsupervised depth estimation model. Specifically, we apply a reliable data augmentation mechanism to minimize the loss of the disparity map generated by the original image and the augmented image, respectively, which will enhance the robustness of the image in the prediction of color fluctuation. At the same time, we aggregate the features of different layers extracted by a pre-trained VGG16 network to explore the higher-level perceptual differences between the input image and the generated one. Ablation studies demonstrate the effectiveness of each components, and PDANet shows high-quality depth estimation results on the KITTI benchmark, which optimizes the state-of-the-art method from 0.114 to 0.084, measured by absolute relative error for depth estimation.
Prostate tumor segmentation from multi-modality magnetic resonance (MR) images is indispensable for diagnosis and treatment of prostate cancer. Existing methods typically over-look the specific distribution of prostate tumors in MR images and also the efficacy of extracting features from different MR modalities in the tumor segmentation task. In this paper, we address these limitations by proposing a novel localization-to-segmentation framework. First, we design a localization stage to locate tumor slices precisely via a label order consistency (LOC) strategy, by explicitly utilizing tumor distribution prior. Then, in the subsequent segmentation stage, we develop an attention-based multi-modality collaborative learning (MCL) module, to extract high-level modality-specific features while focusing on aggregating complementary features across modalities, for segmenting tumors from the localized tumor slices. Experimental results demonstrate that our method achieves state-of-the-art segmentation performance on an in-house prostate MRI dataset, especially for tumors with low contrast to the surrounding tissues.