For autonomous driving, it is important to detect obstacles in all scales accurately for safety consideration. In this paper, we propose a new spatial attention fusion (SAF) method for obstacle detection using mmWave radar and vision sensor, where the sparsity of radar points are considered in the proposed SAF. The proposed fusion method can be embedded in the feature-extraction stage, which leverages the features of mmWave radar and vision sensor effectively. Based on the SAF, an attention weight matrix is generated to fuse the vision features, which is different from the concatenation fusion and element-wise add fusion. Moreover, the proposed SAF can be trained by an end-to-end manner incorporated with the recent deep learning object detection framework. In addition, we build a generation model, which converts radar points to radar images for neural network training. Numerical results suggest that the newly developed fusion method achieves superior performance in public benchmarking. In addition, the source code will be released in the GitHub.

Feature selection (FS) is a key factor for the performance of machine learning algorithms, as not all data and hence features are related to the various tasks. In this paper, we propose a novel scheme for convolutional FS for machine learning algorithms in computer vision. As not all the convolutional features are related to visual tracking, removing the unrelated ones will dramatically reduce the complexity and improve the algorithm performance. However, how to identify and select features related to the visual tracking task is still a challenge for machine learning algorithms. In the proposed scheme, a novel adaptive weights-objective function approach is established to evaluate and select the features. Furthermore, a quadratic programming method is introduced which improves the optimization efficiency. The experimental results demonstrate that our proposed scheme achieves superior performance compared to the state-of-art trackers on the challenging benchmarks in computer vision.

Image colorization has a widespread application in video and image restoration in the past few years. Recently, automatic colorization methods based on deep learning have shown impressive performance. However, these methods map grayscale image input into multi-channel output directly. In the process, it usually loses detailed information during feature extraction, resulting in abnormal colors in local areas of the colorization image. To overcome abnormal colors and improve colorization quality, we propose a novel Double-Channel Guided Generative Adversarial Network (DCGGAN). It includes two modules: a reference component matching module and a double-channel guided colorization module. The reference component matching module is introduced to select suitable reference color components as auxiliary information of the input. The double-channel guided colorization module is designed to learn the mapping relationship from the grayscale to each color channel with the assistance of reference color components. Experimental results show that the proposed DCGGAN outperforms existing methods on different quality metrics and achieves state-of-the-art performance.

Nowadays, faces in videos can be easily replaced with the development of deep learning, and these manipulated videos are realistic and cannot be distinguished by human eyes. Some people maliciously use the technology to attack others, especially celebrities and politicians, causing destructive social impacts. Therefore, it is imperative to design an accurate method for detecting face manipulation. However, most of the existing methods adopt single convolutional neural network as the feature extraction module, causing the extracted features to be inconsistent with the human visual mechanism. Moreover, the rich details and semantic information cannot be reflected with single feature, limiting the detection performance. Therefore, this paper tackles the above problems by proposing a novel face manipulation detection method based on a supervised multi-feature fusion attention network (SMFAN). Specifically, the capsule network is used for face manipulation detection, and the SMFAN is added to the original capsule network to extract details of the fake face image. Further, the focal loss is used to realize hard example mining. Finally, the experimental results on the public dataset FaceForensics++ show that the proposed method has better performance.

Learning from label proportions (LLP) is a new kind of learning problem which has attracted wide interest in machine learning. Different from the well-known supervised learning, the training data of LLP is in the form of bags and only the proportion of each class in each bag is available. Actually, many modern applications can be successfully abstracted to this problem such as modeling voting behaviors and spam filtering. However, time-consuming training is still a challenge for LLP, which becomes a bottleneck especially when addressing large bags and bag sizes. In this paper, we propose a fast algorithm called multi-class learning from label proportions by extreme learning machine (LLP-ELM), which takes advantage of an extreme learning machine with fast learning speed to solve multi-class learning from label proportions. Firstly, we reshape the hidden layer output matrix and the training data target matrix of an extreme learning machine to adapt to the proportion information instead of the real labels. Secondly, a robust loss function with a regularization term is formulated and two efficient solutions are provided to different cases. Finally, various experiments demonstrate the significant speed-up of the proposed model with better accuracies on different datasets compared with several state-of-the-art methods.