Flying insects demonstrate exceptional proficiency in detecting and pursuing conspecifics and prey within a cluttered environment, inspiring the development of computational models for small object detection. While existing bioinspired models are dedicated to resolving small moving instead of stationary object detection, few studies have systematically explored the role of visual motion in detection. Here, we developed a fly-inspired model on the basis of the hypothesis that combining visual motion features and luminance features is critical for small moving object detection. We thoroughly investigated the effect of feature combination under diverse stimulus conditions. Simulations indicated that the model exhibited hyperacute object detection, a capability not generally believed to emerge on the basis of motion detection. When tested with a moving background in realistic scenarios, the model demonstrated enhanced efficiency and robustness relative to models relying solely on luminance features. This enhancement was independent of whether visual motion was extracted by two- or three-arm motion detectors. The results suggested that small object detectors within the visual systems of flying insects could be optimally tuned to utilize the limited features inherent to tiny objects.

Optical flow is pivotal in video-based tasks, yet existing methods mostly focus on medium-/large-size objects, while underperforming when characterizing the motion of tiny objects. To bridge this gap, we introduce the On-off Time-delay with Hassenstein-Reichardt correlator (OTHR), a computationally efficient scheme inspired by the primate visual cortex's direction selectivity mechanism. OTHR kernels, applied across multiple frames, discern bright/dark luminance changes along a specific direction over a time delay, effectively estimating motion of tiny objects amidst noise and static backgrounds. Notably, OTHR integrates seamlessly with leading deep learning flow estimation models such as RAFT and FlowFormer. We also propose refined evaluation metrics for tiny objects and contribute a new dataset featuring such objects to aid algorithm development. Our experiments confirm OTHR's superiority over competing methods, particularly in enhancing state-of-the-art models' performance on tiny object motion estimation at minimal cost. Specifically, for objects less than 100 pixels, OTHR reduces RAFT and FlowFormer's errors by 22.03% and 83.50%, respectively. The codes will be accessible at https://github.com/JaneEliot/OTHR.

In order to tackle the challenges associated with low detection performance and high computational resource consumption, which result from diverse scenes, complex backgrounds, small and dense targets, and large data volumes in remote sensing images, we propose an improvement on real-time detection transformer (RT-DETR) for detecting objects in remote sensing images. By introducing the Mosaic9 data augmentation technique, we enhance the model’s adaptability to multi-scene targets, as well as its target recognition performance in different backgrounds. Furthermore, to decrease resource usage without compromising detection precision, we replace the Basic Block structure in the backbone of RT-DETR with the more lightweight FasterNet Block. Finally, we enhance the adaptive intra-scale feature interaction (AIFI) by replacing multi-head self-attention (MHSA) with deformable attention, enabling the model to dynamically adjust its attention range and better capture distinctive target features in complex scenarios. Experimental validation conducted on the DSTD dataset reveals that the modified RT-DETR model achieves a detection accuracy of 94.9%, representing an improvement of 1% compared to the baseline RT-DETR. Simultaneously, the improvements reduce GFLOPs, total parameters, and overall model size by approximately 12.4%, 15.5%, and 15.2%, respectively, thus realizing a balance between performance and lightweight architecture. Moreover, generalization experiments on the NWPU VHR-10 dataset further substantiate the enhanced model’s robustness and adaptability across diverse remote sensing scenes, confirming the efficacy and practical value of the proposed enhancements.

Accurate multiple sequence alignment (MSA) is imperative for the comprehensive analysis of biological sequences. However, a notable challenge arises as no single MSA tool consistently outperforms its counterparts across diverse datasets. Users often have to try multiple MSA tools to achieve optimal alignment results, which can be time-consuming and memory-intensive. While the overall accuracy of certain MSA results may be lower, there could be local regions with the highest alignment scores, prompting researchers to seek a tool capable of merging these locally optimal results from multiple initial alignments into a globally optimal alignment. In this study, we introduce Two Pointers Meta-Alignment (TPMA), a novel tool designed for the integration of nucleic acid sequence alignments. TPMA employs two pointers to partition the initial alignments into blocks containing identical sequence fragments. It selects blocks with the high sum of pairs (SP) scores to concatenate them into an alignment with an overall SP score superior to that of the initial alignments. Through tests on simulated and real datasets, the experimental results consistently demonstrate that TPMA outperforms M-Coffee in terms of aSP, Q, and total column (TC) scores across most datasets. Even in cases where TPMA's scores are comparable to M-Coffee, TPMA exhibits significantly lower running time and memory consumption. Furthermore, we comprehensively assessed all the MSA tools used in the experiments, considering accuracy, time, and memory consumption. We propose accurate and fast combination strategies for small and large datasets, which streamline the user tool selection process and facilitate large-scale dataset integration. The dataset and source code of TPMA are available on GitHub (https://github.com/malabz/TPMA).

Recently, the transformer has enabled the speed-oriented trackers to approach state-of-the-art (SOTA) performance with high-speed thanks to the smaller input size or the lighter feature extraction backbone, though they still substantially lag behind their corresponding performance-oriented versions. In this paper, we demonstrate that it is possible to narrow or even close this gap while achieving high tracking speed based on the smaller input size. To this end, we non-uniformly resize the cropped image to have a smaller input size while the resolution of the area where the target is more likely to appear is higher and vice versa. This enables us to solve the dilemma of attending to a larger visual field while retaining more raw information for the target despite a smaller input size. Our formulation for the non-uniform resizing can be efficiently solved through quadratic programming (QP) and naturally integrated into most of the crop-based local trackers. Comprehensive experiments on five challenging datasets based on two kinds of transformer trackers, \ie, OSTrack and TransT, demonstrate consistent improvements over them. In particular, applying our method to the speed-oriented version of OSTrack even outperforms its performance-oriented counterpart by 0.6\% AUC on TNL2K, while running 50\% faster and saving over 55\% MACs. Codes and models are available at https://github.com/Kou-99/ZoomTrack.

Building high-resolution lithological models using seismic data can facilitate decision-makings for earth resources development, but they are subject to considerable uncertainties due to the limited seismic resolution. Traditionally, high-resolution lithological models are built deterministically or stochastic simulation is performed using seismic as “soft data” in geostatistical contexts. Many approaches have been developed to do the above, but in this work, we explicitly account for the uncertain relationship between seismic data and lithological models. We introduce a data-driven Bayesian approach to first quantify soft data uncertainty by generating multiple lithology proportions from seismic. The spatial lithology proportion uncertainty arises from the uncertain relationship between low-resolution seismic and lithology proportions. We quantify this proportion uncertainty by learning the statistical relationships with seismic using high-resolution borehole data. Once conditioning the statistical relationships to observed seismic, we can generate multiple realizations of spatial proportions. With the generated proportions as volume averages, a sequential indicator simulation (SISim) is then performed to build high-resolution lithofacies models. This approach is applied to a real channelized turbidite system with four lithofacies. We demonstrate that the generated high-resolution lithological models can preserve the global uncertainties captured by proportion trends while locally matching borehole observations. When compared with the conventional approaches that use deterministic soft data, our statistical learning approach avoids the problem of underestimating reservoir model uncertainty. More importantly, the lithofacies models from the proposed method are less likely to be falsified by observed data. Additionally, an open-source Python library for the uncertainty quantification is provided.