Xiang Li

Object detection from aerial platforms under adverse atmospheric conditions, particularly haze, is paramount to robust drone autonomy. Yet, this domain remains largely underexplored, primarily hindered by the absence of specialized benchmarks. To bridge this gap, we present HazyDet, the first, large-scale benchmark specifically designed for drone-view object detection in hazy conditions. Comprising 383,000 real-world instances derived from both naturally hazy captures and synthetically hazed scenes augmented from clear images, HazyDet provides a challenging and realistic testbed for advancing detection algorithms. To address the severe visual degradation induced by haze, we propose the Depth-Conditioned Detector (DeCoDet), a novel architecture that integrates a Depth-Conditioned Kernel to dynamically modulate feature representations based on depth cues. The practical efficacy and robustness of DeCoDet are further enhanced by its training with a Progressive Domain Fine-Tuning (PDFT) strategy to mitigate synthetic-to-real domain shifts, and a Scale-Invariant Refurbishment Loss (SIRLoss) to ensure resilient learning from potentially noisy depth annotations. Comprehensive empirical validation on HazyDet substantiates the superiority of our unified DeCoDet framework, which achieves state-of-the-art performance, surpassing the closest competitor by a notable +1.0% mAP on challenging real-world hazy test scenarios. Our dataset and toolkit are available at https://github.com/GrokCV/HazyDet.

With the rapid development of panoramic photography, panoramic depth perception has become a critical aspect of 3D computer vision. Different from perspective images, panoramas provide 360° scene context whilst introducing latitude-wise distortion, making it difficult for conventional convolutions (e.g., $3\times 3$ ) to effectively deal with. In this paper, we propose a Large Distortion-aware Kernel (LDK), which allows convolutions to handle distortion and capture rich panoramic context simultaneously. Specifically, LDK divides the feature into multiple groups based on distortion degree and applies very large kernel convolutions that adapt their horizontal receptive fields to the distortion degree as well. This design enables the convolutions to directly cover distorted regions while incorporating more scene context. In addition, considering that the large kernels only have local and long-range vision, LDK then propagates the 360° context via global operator to compensate for global semantic cues. Interestingly, we find that the local, long-range, and global operators can be skillfully integrated while maintaining accuracy. Finally, we stack multiple LDK units into a deep network, PanoKernel. Without bells and whistles, PanoKernel outperforms existing SoTA methods even with lower model complexity on three panoramic depth perception tasks.

New knowledge originates from the old. The various types of elements, deposited in the training history, are a large amount of wealth for improving learning deep models. In this survey, we comprehensively review and summarize the topic–“Historical Learning: Learning Models with Learning History”, which learns better neural models with the help of their learning history during its optimization, from three detailed aspects: Historical Type (what), Utility Part (where) and Storage Form (how). To our best knowledge, it is the first survey that systematically studies the methodologies which make use of various historical statistics when training deep neural networks. The discussions with related topics like recurrent/memory networks, ensemble learning, and reinforcement learning are demonstrated. We also expose future challenges of this topic and encourage the community to pay attention to the think of historical learning principles when designing algorithms. The paper list related to historical learning is available at https://github.com/Martinser/Awesome-Historical-Learning.