Cracks can provide early warning for damage of roadway pavement and civil infrastructure. However, due to the diversity of texture and color of the materials, the complexity of the background, and the interference factors such as noise, low illumination, and tiny cracks, crack detection using image sensor is extremely challenging. To address the above problems, we propose a surface crack detection method based on semantic segmentation network. Firstly, to cope with the interference of various complex situations in wild scenes, we propose an attention mechanism based on group-wise normalization to suppress meaningless features and highlight meaningful features. We compress feature maps into components of a vector along the channel dimension and normalize these components by each group in the vector, then use the learnable parameters in group-wise normalization to weight feature maps by channel. Secondly, to enlarge the receptive field, we propose a "padding-shifting-cutting" convolution. Our proposed convolution cyclically shifts the feature maps by different count of pixels horizontally and vertically. After shifted, each part of the feature map exchanges information, thus our model can obtain longer range crack information and detect tiny cracks better. Finally, experimental results show that our method has achieved better performance than other methods on four public datasets: CFD, CrackTree260, DeepCrack, and CrackLS315. Our model surpasses most excellent existing models in terms of Precision, Recall, F1-score, mIoU, and IoU on the CrackTree260 dataset.

We propose a novel optimization framework for video object segmentation, given the initial annotations of objects in the keyframes of an input video sequence. In this work, video data is represented by a Markov Random Field model, and segmentation is achieved by finding the minimum graph cut label assignment. More specifically, we first create a bilateral representation of the input video sequence which reduces the size of the graph that the min-cut must operate on. We then introduce dynamic appearance models to learn the segmentation likelihoods, and the reliability of likelihoods is measured to identify false likelihoods that may cause segmentation errors. Thus, the model accurately describes changes in the object’s appearance that have evolved over time. Furthermore, we augment spatial and temporal connections using a soft higher-order potential, ensuring long-range label consistency in the segmentation. We provide extensive analysis and evaluation with respect to the influence of each component of the framework through the ablation study. Experiments on three benchmark datasets (DAVIS 2016, YouTube-Objects and SegTrack v2) show that our method achieves competitive performance compared to state-of-the-art while having the order of magnitude faster runtime.