Perceptual quality assessment of the videos acquired in the wilds is of vital importance for quality assurance of video services. The inaccessibility of reference videos with pristine quality and the complexity of authentic distortions pose great challenges for this kind of blind video quality assessment (BVQA) task. Although model-based transfer learning is an effective and efficient paradigm for the BVQA task, it remains to be a challenge to explore what and how to bridge the domain shifts for better video representation. In this work, we propose to transfer knowledge from image quality assessment (IQA) databases with authentic distortions and large-scale action recognition with rich motion patterns. We rely on both groups of data to learn the feature extractor and use a mixed list-wise ranking loss function to train the entire model on the target VQA databases. Extensive experiments on six benchmarking databases demonstrate that our method performs very competitively under both individual database and mixed databases training settings. We also verify the rationality of each component of the proposed method and explore a simple ensemble trick for further improvement.

Radiography is an essential basis for the diagnosis of fractures. For the pediatric elbow joint diagnosis, the doctor needs to diagnose abnormalities based on the location and shape of each bone, which is a great challenge for AI algorithms when interpreting radiographs. Bone instance segmentation is an effective upstream task for automatic radiograph interpretation. Pediatric elbow bone instance segmentation is a process by which each bone is extracted separately from radiography. However, the arbitrary directions and the overlapping of bones pose issues for bone instance segmentation. In this paper, we design a detection-segmentation pipeline to tackle these problems by using rotational bounding boxes to detect bones and proposing a robust segmentation method. The proposed pipeline mainly contains three parts: (i) We use Faster R-CNN-style architecture to detect and locate bones. (ii) We adopt the Oriented Bounding Box (OBB) to improve the localizing accuracy. (iii) We design the Global-Local Fusion Segmentation Network to combine the global and local contexts of the overlapped bones. To verify the effectiveness of our proposal, we conduct experiments on our self-constructed dataset that contains 1274 well-annotated pediatric elbow radiographs. The qualitative and quantitative results indicate that the network significantly improves the performance of bone extraction. Our methodology has good potential for applying deep learning in the radiography’s bone instance segmentation.

No-reference image quality assessment (NR-IQA), which devotes to predicting image quality without relying on the corresponding pristine counterpart, develops rapidly in recent years. However, little investigation has been dedicated to quality assessment of realistic night-time images. Existing NR-IQA algorithms laboriously cope with this night-time scenario since complicated authentic distortions such as low contrast, blurred details, and reduced visibility usually appear on it. In this paper, we propose an end-to-end NR-IQA model to meet this challenge based on a multi-stream deep convolutional neural network (DCNN). Two streams, brightness-aware CNN and naturalness-aware CNN are constructed respectively by a brightness-altered image identification task with a self-established dataset and a quality-prediction regression task with an existing authentically-distorted IQA dataset to improve quality-aware initializations. In this case, given the quick convergence and little transformation in the lower layers, a shallow-layer-shared architecture is explored to reduce computational cost. Finally, the features of these two pipelines are collected by an effective pooling method and then concatenated as the image representation for fine-tuning. The effectiveness and efficiency of the proposed method are verified by several different experiments on the NNID, CCRIQ and LIVE Challenge databases. Furthermore, the superiority of wide applications such as for contrast-distorted and driving scenarios is demonstrated on the CID2013, CCID2014 and BBD-100k databases.

Over the past few decades, video quality assessment (VQA) has become a valuable research field. The perception of in-the-wild video quality without reference is mainly challenged by hybrid distortions with dynamic variations and the movement of the content. In order to address this barrier, we propose a no-reference video quality assessment (NR-VQA) method that adds the enhanced awareness of dynamic information to the perception of static objects. Specifically, we use convolutional networks with different dimensions to extract low-level static-dynamic fusion features for video clips and subsequently implement alignment, followed by a temporal memory module consisting of recurrent neural networks branches and fully connected (FC) branches to construct feature associations in a time series. Meanwhile, in order to simulate human visual habits, we built a parametric adaptive network structure to obtain the final score. We further validated the proposed method on four datasets (CVD2014, KoNViD-1k, LIVE-Qualcomm, and LIVE-VQC) to test the generalization ability. Extensive experiments have demonstrated that the proposed method not only outperforms other NR-VQA methods in terms of overall performance of mixed datasets but also achieves competitive performance in individual datasets compared to the existing state-of-the-art methods.

The Markov random field (MRF) method is widely used in remote sensing image semantic segmentation because of its excellent spatial (relationship description) ability. However, there are some targets that are relatively small and sparsely distributed in the entire image, which makes it easy to misclassify these pixels into different classes. To solve this problem, this paper proposes an object-based Markov random field method with partition-global alternately updated (OMRF-PGAU). First, four partition images are constructed based on the original image, they overlap with each other and can be reconstructed into the original image; the number of categories and region granularity for these partition images are set. Then, the MRF model is built on the partition images and the original image, their segmentations are alternately updated. The update path adopts a circular path, and the correlation assumption is adopted to establish the connection between the label fields of partition images and the original image. Finally, the relationship between each label field is constantly updated, and the final segmentation result is output after the segmentation has converged. Experiments on texture images and different remote sensing image datasets show that the proposed OMRF-PGAU algorithm has a better segmentation performance than other selected state-of-the-art MRF-based methods.