Label assignment refers to determining positive/negative labels for each sample to supervise the training process. Existing Siamese-based trackers primarily use fixed label assignment strategies according to human prior knowledge; thus, they can be sensitive to predefined hyperparameters and fail to fit the spatial and scale variations of samples. In this study, we first develop a novel dynamic label assignment (DLA) module to handle the diverse data distributions and adaptively distinguish the foreground from the background based on the statistical characteristics of the target in visual object tracking. The core of DLA module is a two-step selection mechanism. The first step selects candidate samples according to the Euclidean distance between training samples and ground truth, and the second step selects positive/negative samples based on the mean and standard deviation of candidate samples. The proposed approach is general-purpose and can be easily integrated into anchor-based and anchor-free trackers for optimal sample-label matching. According to extensive experimental findings, Siamese-based trackers with DLA modules can refine target locations and outperform baseline trackers on OTB100, VOT2019, UAV123 and LaSOT. Particularly, DLA-SiamRPN++ improves SiamRPN++ by 1% AUC and DLA-SiamCAR improves SiamCAR by 2.5% AUC on OTB100. Furthermore, hyper-parameters analysis experiments show that DLA module hardly increases spatio-temporal complexity, the proposed approach maintains the same speed as the original tracker without additional overhead.

Small-object detection has long been a challenge. High-megapixel cameras are used to solve this problem in industries. However, current detectors are inefficient for high-resolution images. In this work, we propose a new module called Pre-Locate Net, which is a plug-and-play structure that can be combined with most popular detectors. We inspire the use of classification ideas to obtain candidate regions in images, greatly reducing the amount of calculation, and thus achieving rapid detection in high-resolution images. Pre-Locate Net mainly includes two parts, candidate region classification and behavior classification. Candidate region classification is used to obtain a candidate region, and behavior classification is used to estimate the scale of an object. Different follow-up processing is adopted according to different scales to balance the variance of the network input. Different from the popular candidate region generation method, we abandon the idea of regression of a bounding box and adopt the concept of classification, so as to realize the prediction of a candidate region in the shallow network. We build a high-resolution dataset of aircraft and landing gears covering complex scenes to verify the effectiveness of our method. Compared to state-of-the-art detectors (e.g., Guided Anchoring, Libra-RCNN, and FASF), our method achieves the best mAP of 94.5 on 1920 × 1080 images at 16.7 FPS.