Learning a robust object detector in adverse weather with real-time efficiency is of great importance for the visual perception task for autonomous driving systems. In this paper, we propose a framework to improve the YOLO to a robust detector, denoted as R(obust)-YOLO, without the need for annotations in adverse weather. Considering the distribution gap between the normal weather images and the adverse weather images, our framework consists of an image quasi-translation network (QTNet) and a feature calibration network (FCNet) for adapting the normal weather domain to the adverse weather domain gradually. Specifically, we use the simple yet effective QTNet for generating images that inherit the annotations in the normal weather domain and interpolate the gap between the two domains. Then, in FCNet, we propose two kinds of adversarial learning-based feature calibration modules to effectively align the feature representations in two domains in a local-to-global manner. With such a learning framework, our R-YOLO does not change the original YOLO structure, and thus it is applicable to all YOLO-series detectors. Extensive experimental results of our R-YOLOv3, R-YOLOv5, and R-YOLOX on both hazy and rainy datasets show that our method outperforms other detectors with dehaze/derain as the pre-processing step and other unsupervised domain adaptation-based detectors, which confirms the effectiveness of our method on improving the robustness by only leveraging the unlabeled adverse weather images. Our code and pre-trained models are available at: https://github.com/qinhongda8/R-YOLO.

Previous methods in salient object detection (SOD) mainly focused on favorable illumination circumstances while neglecting the performance in low-light condition, which significantly impedes the development of related down-stream tasks. In this work, considering that it is impractical to annotate the large-scale labels for this task, we present a framework (HDNet) to detect the salient objects in low-light images with the synthetic images. Our HDNet consists of a foreground highlight sub-network (HNet) and an appearance-aware detection sub-network (DNet), both of which can be learned jointly in an end-to-end manner. Specifically, to highlight the foreground objects, we design the HNet to estimate the parameters to adjust the dynamic range for each pixel adaptively, which can be trained via the weak supervision signals of the salient object labels. In addition, we design a simple detection network (DNet) with a contextual feature fusion module and a multi-scale feature refine module for detailed feature fusion and refinement. Furthermore, we contribute the first annotated dataset for salient object detection in low-light images (SOD-LL), including 6,000 labeled synthetic images (SOD-LLS) and 2,000 labeled real images (SOD-LLR). Experimental results on SOD-LL and other low-light videos in the wild demonstrate the effectiveness and generalization ability of our method. Our dataset and code are available at https://github.com/Ylinyuan/HDNet.

This paper presents a new approach of compensating the zero-drift error of electronic balance based on automatic step adjustment and least mean square method (Autostep-LMS), and the detailed algorithm of Autostep-LMS is given. Unlike the classical variable step-size LMS algorithms, such as Aboulnasr-LMS algorithm, Versoria-function-LMS algorithm and Gradient-LMS algorithm, the step control parameters of Autostep-LMS algorithm are automatically adjusted, which is helpful to reduce the complexity of how to adjust these algorithm parameters. By comparing with these classical variable step-size LMS algorithms, the simulation results show that the Autostep-LMS algorithm has faster convergence speed and smaller steady-state error. This Autostep-LMS algorithm is applied to the electronic balance which range is 200g and the resolution is 1mg, the experimental results show that the zero-drift error, repeatability error and nonlinear error of the electronic balance are all less than 5mg, which is superior to the recommendation of the class II electronic balance.

Unsupervised domain adaptation (UDA) is an effective scheme to improve the performance of an object detector in foggy scenes by adapting labeled normal images (source domain) to unlabeled foggy images (target domain). Existing methods leverage the Teacher-Student mutual learning framework, i.e., unbiased mean teacher, to produce pseudo-labels of the foggy images for self-training. This kind of general-purpose design cannot fully explore the characteristics of foggy images, resulting in the low-quality pseudo-labels which degrades the mutual learning between the teacher and student network. Addressing this problem, we propose a foggy-aware teacher (FAT) framework that improves the quality of pseudo-labels by exploring the fog characteristics from two aspects: 1) we design a light-weight foggy-aware module (FAM), which is inserted before the teacher network to estimate the fog information, i.e., the global atmospheric light, and produce the source-like (defogged) images to mitigate the domain shift. 2) we propose a foggy-information-guided image fusion (FIF) scheme, which is inserted before the student network to produce the interpolated images by employing the fog characteristics on the normal images. The student can be trained with the high-quality pseudo-labels produced by the teacher and the ground-truth labels supplied by the normal images. The entire mutual learning framework can then form a virtuous self-training cycle to improve the performance on foggy images. Extensive experiments demonstrate that our FAT framework is effective in improving detector performance under foggy conditions, and has achieved better results than state-of-the-art methods on benchmark datasets Foggy Cityscapes, RTTS, and VOC_foggy.

This paper presents a method for compensating the load cell's nonlinear error based on neural network with the second derivative constraints. In this method, for improving the ability of the neural network in the case of lack of samples, the prior knowledge of load cell, i.e., the second derivative of the load cell's input-output function is less than zero, is used to construct the constraint of neural network. The augmented performance index of training neural network with penalty function method is founded, and its detailed training algorithm is given. In addition, the performance of neural network affected by the punishing factor is discussed. This proposed method can reduce the NN's generalization error when the training samples are insufficient. The experimental results show that the generalization ability of this proposed neural network is better than that of the conventional data inducing neural network (DINN, i.e. training neural network by only using data samples and not any constraints), and the nonlinear error of load cell with this proposed method is far less than that of DINN.

As an important task in autonomous driving of intelligent vehicles, the autonomous parking technology has a broad market potential. The traditional parking algorithm based on trajectory planning often results in significant tracking errors due to the non-linearity of vehicle dynamics. While general reinforcement learning algorithms may require the assistance of domain experts due to the complexity of the parking scene. However, relying on expert demonstrations can also be problematic due to the high collection cost and potential for uneven quality of the demonstrations. To address these problems, we propose an end-to-end reinforcement learning-based parking method using expert demonstrations. Firstly, we add Gaussian noise to the collected expert demonstrations for data augmentation, which expands the diversity of samples and guides the agent to learn the expert policy quickly. Then we introduce the mixed priority sampling strategy, and employ it on the augmented expert demonstrations and experience obtained from exploration in the environment to jointly train the agent. We evaluate our method on the benchmark algorithm, Soft Actor-Critic, on the CARLA autonomous driving simulation platform with vertical and parallel parking environments. The experimental results indicate that the proposed algorithm can greatly improve the convergence rate and achieve a higher parking success rate.