Optical flow estimation is a low-level module in computer vision, widely used in tasks such as visual odometry, autonomous driving, high dynamic range (HDR) imaging, and action recognition. Existing event-based optical flow estimation approaches suffer from scarcity of dense real-world datasets, while some unsupervised frameworks have reduced reliance on large-scale datasets by forward-backward consistency loss, they primarily exploit a 1D temporal reversal, while largely ignoring the rotational and scaling motions ubiquitous in robotics and automotive scenes. This work introduces radial consistency, a self-supervised pre-training framework that maps the event stream to log-polar coordinates and tessellates the spatial domain into K radial rings and L angular sectors, a shared encoder-decoder to predict four complementary flow fields whose cyclic sum is driven to zero, yielding a closed-loop constraint that generalizes the classical forward-backward check to 360° within a sector. Our core contribution, the radial consistency loss, is completely label-free, together with auxiliary terms, enabling self-supervised pre-training on large-scale event data. We optionally apply supervised fine-tuning on small labeled sets to adapt to specific domains, achieving competitive accuracy with fully supervised methods. Validation experiments on Multi Vehicle Stereo Event Camera (MVSEC) dataset demonstrate strong performance: our method achieves 0.67 EPE averaged across all sequences, surpassing E-RAFT (0.89 EPE) and EV-FlowNet (1.10 EPE), without any additional data. On the rotation-heavy indoor_flying3 sequence specifically, we achieve 0.93 EPE (fine-tuned) and 1.49 EPE (self-supervised only) vs. E-RAFT 1.66. We also improve upon E-RAFT in computational efficiency [55 frames per second (FPS) and 26 giga floating-point operations (GFLOPs) vs. 42 FPS and 38 GFLOPs], while requiring only minimal supervised fine-tuning.
Robust 3D reconstruction is essential for robotics and embodied perception. Recent feed-forward approaches such as DUSt3R have demonstrated impressive progress in dense 3D reconstruction from RGB images, achieving global geometric consistency and strong generalization. However, extending such dense 3D reconstruction to event cameras remains challenging due to their asynchronous, sparse, and highly dynamic nature, as well as the lack of large-scale, well-labeled datasets. In this work, we introduce Event3R, a feed-forward framework that directly maps asynchronous event streams to globally consistent 3D point clouds. Event3R represents incoming events as spatial-temporal voxels, enabling time-aware feature integration through a temporal attention module that enhances the module's temporal feature learning. To further strengthen temporal representation learning and reduce reliance on labeled data, we propose a Masked Bin Modeling (MBM) strategy for self-supervised pre-training, enabling robust temporal representation learning with minimal labeled data, and retain it as an auxiliary fine-tuning objective. In addition, contrastive alignment and consistency regularization losses are incorporated during fine-tuning to reinforce structural correspondence and temporal coherence across views. Extensive experiments on both synthetic and real-world benchmarks demonstrate that Event3R achieves robust, temporally consistent, and globally aligned 3D reconstructions, significantly outperforming existing event-based methods.
Jian Huang, Haotian Shen, Xinhao Lou et al.· 0 citations
Motion-only multiobject tracking (MOT) suffers from ID switches in uniform-appearance and deformation-heavy scenes. In these settings, appearance cues become less reliable, so stable identities depend mainly on motion information. Existing methods often process all bounding-box variables together, which can weaken cues needed for prediction and matching. We address this problem by treating each trajectory as a heterogeneous multivariate time series (MTS) and redesigning the motion-only pipeline for embedding, encoding, and matching. HSMTrack separates box-state variables before modeling their temporal and cross-variable relationships, then uses deformation-aware matching for identity association. The method requires no appearance branch and can serve as a post-detection motion module in vision-sensor tracking pipelines. Its SSM-based encoder has linear complexity with respect to trajectory length, reducing modeling cost compared with attention-based alternatives. HSMTrack achieves 59.6 IDF1 and 42.9 AssA on DanceTrack, and 77.9 IDF1 and 67.2 AssA on SportsMOT. Under a unified end-to-end protocol, it reaches 34.1 frames/s on RTX 4090 and 10.5 frames/s on Jetson Orin NX.
Tianjing Cheng, Qingyuan Yu, Bo Jiang et al.· IEEE Sensors Journal· 0 citations
Optical flow estimation is a fundamental task in computer vision and visual sensing systems. However, most existing approaches are designed for normal illumination conditions. In low-light scenarios, inherent imaging noise and low contrast lead to noticeable feature degradation and matching ambiguity, which compromise estimation accuracy. To address these issues, this article proposes spatial-frequency dual-domain refinement and motion prior calibration (SFRC)-Flow, a robust low-light optical flow estimation method that integrates spatial–frequency dual-domain feature refinement and motion prior calibration to ensure reliable feature learning and alleviate matching ambiguity. Specifically, the dual-domain refinement encoder (DDRE) first decomposes shallow spatial features into high-frequency local details and low-frequency global structures. Subsequently, we propose the global–local feature alignment module (GLFAM) to resolve spatial and semantic inconsistencies between these decomposed features across different branches via cross-branch feature alignment. Building upon this, we further introduce a cascaded three-stage frequency-domain refinement module (FDRM) to compensate for the limited long-range modeling capability of spatial-only operations and recover degraded motion features. Finally, we present the motion prior-aware calibration module (MPACM) to incorporate motion cues into window-constrained semantic attention. This module produces motion vectors as prior knowledge to calibrate the subsequent flow regression process. Extensive experiments on the flying chairs-dark noise (FCDN), various brightness optical flow (VBOF), and teledyne forward looking infrared advanced driver assistance systems thermal dataset (FLIR ADAS) datasets show that SFRC-Flow achieves competitive accuracy for low-light optical flow estimation.
AD-YOLO is presented, a dual-level framework that tackles pseudo-label noise and limited multi-scale adaptability when applied to semi-supervised object detection frameworks from both the detector architecture and the SSOD pipeline.
Jie Long· Engineering Research Express· 0 citations
Camera-based 3D object detection has attracted widespread attention for autonomous driving applications. However, existing methods often lack effective feature screening mechanisms, resulting in an extremely low spatio-temporal signal-to-noise ratio in complex scenes. Specifically, distracting background projections, feature misalignment caused by dynamic objects, and frequent occlusions jointly lead to severe ambiguity and loss of object features. To alleviate these issues, we propose STRDet, a robust 3D object detection framework based on progressive spatio-temporal feature refinement. First, we propose a Semantic-guided Context Refinement (SCR) module that explicitly suppresses background interference prior to the view transformation, thereby blocking noise propagation. Second, we design the Differential-aware Feature Alignment (DFA) and Residual-based Adaptive Gated Fusion (RAGF) modules, which leverage feature difference maps as motion saliency indicators to guide deformable alignment and employ gating mechanisms to selectively integrate historical motion cues, effectively resolving dynamic alignment failures and mitigating feature loss under occlusion. Extensive experiments on the nuScenes dataset demonstrate that STRDet effectively enhances feature purity and coherence, yielding significant improvements and achieving 46.95\% mAP and 55.37\% nuScenes detection score.
Bing Li, Lie Guo, Longxin Guan et al.· Measurement science and tech...· 0 citations
Accurate depth perception is a cornerstone of autonomous driving, yet LiDAR sensors—the primary source of metric depth—remain costly and operationally complex. In this paper, we propose a cross-modal supervision framework that uses sparse LiDAR depth maps solely during training, enabling camera-only dense depth inference at test time. A ConvNeXt-base encoder with an FPN neck and a lightweight depth decoding head is trained with a log-scale L1 loss, gradient consistency term applied exclusively at valid LiDAR pixels (~0.7% pixel density), and an image-guided edge-aware smoothness loss operating on all pixels, alongside a two-phase backbone freeze-then-unfreeze strategy to stabilize early convergence. Evaluated on a large-scale Korean highway dataset of 64,840 frames, our model achieves AbsRel of 0.0675, RMSE of 3.907 m, and $\delta \lt 1.25$ accuracy of 0.943, demonstrating that ultra-sparse LiDAR supervision is sufficient to train competitive monocular depth estimators.
H. Park, Young Hoon Suh, Sun-Kug Jung et al.· International Conference on...· 0 citations
