Fei Wen

Robust 3D registration is a fundamental problem in computer vision and robotics, where the goal is to estimate the geometric transformation between two sets of measurements in the presence of noise and outlier contamination. Existing robust registration methods are mainly built on either maximum consensus (MC) estimators, which first identify inliers and then estimate the transformation, or M-estimators, which directly optimize a robust objective. However, MC-based methods typically ignore residual magnitudes during inlier selection, while many M-estimators do not explicitly couple inlier/outlier identification with model estimation. Thus, a unified and efficient framework that jointly performs inlier identification and accurate transformation estimation remains desirable for challenging 3D registration. In this work, we introduce a unified truncated-loss based formulation for simultaneous inlier identification and model estimation (SIME) and study it in the context of 3D registration. We show that, compared with MC-based robust fitting, SIME can achieve a lower fitting residual because it incorporates residual magnitudes into the inlier selection process. To solve the resulting nonconvex problem, we develop an alternating minimization (AM) algorithm, and further propose an AM method embedded with semidefinite relaxation (AM-R) to alleviate the difficulty caused by the binary inlier variables. We instantiate the proposed framework for 3D rotation search and rigid point-set registration using quaternion-based formulations. Experimental results on both simulated and real-world registration tasks demonstrate that the proposed methods compare favorably with strong baseline solvers, especially in high noise and extreme outliers. In the synthetic experiments, the proposed methods are evaluated under outlier ratios up to 95% and consistently achieve competitive or better accuracy, with clear advantages in high-noise cases. On 3DMatch, SIME (AM) achieves a mean registration success rate of 91.0%. These results show the potential of SIME for reliable 3D registration in practical robotics, computer vision, and geometric perception applications.

The hardware convergence of Artificial Neural Networks (ANNs) and Spiking Neural Networks (SNNs) is hindered by conflicting computational paradigms: dense tensor parallelism versus asynchronous sparse dynamics. Existing unifications typically rely on inefficient spatial partitioning or mode-reconfigurable datapaths, limiting the flexibility needed by heterogeneous ANN-SNN hybrid models requiring frequent cross-domain interaction. To resolve this, we present NeuroUNI, a unified event-driven multi-core architecture. Unlike partitioned designs, NeuroUNI unifies computation at the primitive level using a novel Five-Tuple Event Model, abstracting both continuous activations and discrete spikes to naturally leverage dynamic sparsity. The architecture features a co-optimized hierarchical Macro-Micro- $\mu $ OP ISA, a superscalar SIMD-based microarchitecture, and a decentralized multi-core synchronization protocol. Validated in TSMC 28nm technology via post-synthesis simulation and on a Xilinx VCU129 FPGA prototype, NeuroUNI demonstrates competitive cross-paradigm efficiency. It achieves $35.0\times $ and $1.21\times $ the ANN energy efficiency of the NVIDIA V100 and EyerissV2, respectively, while delivering $5.7\times $ the SNN throughput of TrueNorth. In a unified mapless navigation workload, NeuroUNI attains 422.6 GOPS/W (ANN) and 190.8GSOPS/W (SNN), outperforming Loihi 1 with $56.7\times $ the throughput and $3.85\times $ the energy efficiency, proving the viability of a primitive/ISA-level unified silicon substrate.

Few-Shot Class-Incremental Learning (FSCIL) aims to continually learn novel classes from limited data while retaining knowledge of previously learned classes. To mitigate catastrophic forgetting, most approaches pre-train a powerful backbone on the base session and keep it frozen during incremental sessions. Within this framework, existing studies primarily focus on representation learning in FSCIL, particularly Self-Supervised Contrastive Learning (SSCL), to enhance the transferability of representations and thereby boost model generalization to novel classes. However, they face a trade-off dilemma: improving transferability comes at the expense of discriminability, precluding simultaneous high performance on both base and novel classes. To address this issue, we propose BR-FSCIL, a representation learning framework for the FSCIL scenario. In the pre-training stage, we first design a Hierarchical Contrastive Learning (HierCon) algorithm. HierCon leverages label information to model hierarchical relationships among features. In contrast to SSCL, it maintains strong discriminability when promoting transferability. Second, to further improve the model’s performance on novel classes, an Alignment Modulation (AM) loss is proposed that explicitly facilitates learning of knowledge shared across classes from an inter-class perspective. Building upon the hierarchical discriminative structure established by HierCon, it additionally improves the model’s adaptability to novel classes. Through optimization at both intra-class and inter-class levels, the representations learned by BR-FSCIL achieve a balance between discriminability and transferability. Extensive experiments on mini-ImageNet, CIFAR100, and CUB200 demonstrate the effectiveness of our method, which achieves final session accuracies of 53.83%, 53.04%, and 62.60%, respectively.

Time-of-Flight (ToF) imaging systems are susceptible to various noise and degradation, which can severely affect image quality. Traditional sequential imaging pipelines often suffer from error accumulation due to separate multi-stage processing. Existing end-to-end methods typically rely on noisy-clean depth image pairs for supervised learning. However, acquiring ground-truth is challenging in real-world scenarios due to factors such as Multi-Path Interference (MPI), phase wrapping, and complex noise patterns. In this paper, we propose a self-supervised learning framework for end-to-end ToF imaging, which does not require any noisy-clean pairs yet generalizes well across various off-the-shelf cameras. Our framework leverages the cross-modal dependencies between RGB and depth data as implicit supervision to effectively suppress noise and maintain image fidelity. Additionally, the loss function integrates the statistical characteristics of raw measurement data, enhancing robustness against noise and artifacts. Extensive experiments on both synthetic and real-world data demonstrate that our approach achieves performance comparable to supervised methods, without requiring paired noisy-clean data for training. Furthermore, our method consistently delivers strong performance across all evaluated cameras, highlighting its generalization capabilities. The code is available at https://github.com/WeihangWANG/RGBD_imaging.

Recently, learning-based robotic navigation systems have gained extensive research attention and made significant progress. However, the diversity of open-world scenarios poses a major challenge for the generalization of such systems to practical scenarios. Specifically, learned systems for scene measurement and state estimation tend to degrade when the application scenarios deviate from the training data, resulting in unreliable depth and pose estimation. Toward addressing this problem, this work aims to develop a visual odometry system that can rapidly adapt to diverse novel environments in an online manner. To this end, we construct a self-supervised online adaptation framework where the visual odometry system and the depth estimation module reinforce each other through a continuous measure-calibrate-remeasure cycle. Firstly, we design a monocular depth estimation network with lightweight refiner modules, which enables efficient online adaptation. Then, we construct an objective for self-supervised learning of the depth estimation module based on the output of the visual odometry system and the contextual semantic information of the scene. Specifically, a sparse depth densification module and a dynamic consistency enhancement module are proposed to leverage camera poses and contextual semantics to generate pseudo-depths and valid masks for the online adaptation. Finally, we demonstrate the robustness and generalization capability of the proposed method in comparison with state-of-the-art learning-based approaches on urban, in-house datasets and a robot platform. Code is publicly available at: https://github.com/jixingwu/SOL-SLAM.

Deep learning-based speech enhancement methods are predominantly trained in a supervised manner, relying on synthesized paired noisy-to-clean data. However, acquiring clean speech in real-world scenarios is often difficult or even impractical. To overcome this limitation, we propose a novel unsupervised learning framework for speech enhancement that relies solely on observed noisy speech, called MOS-GAN. Specifically, we leverage generative adversarial networks (GANs), where the generator (the enhancement model) is optimized to maximize the mean opinion score (MOS) guided by a discriminator, while the discriminator (a non-intrusive speech quality metric model) is optimized to predict MOS. However, without using reference clean speech, directly training of MOS-GAN is unstable and cannot achieve satisfactory performance. To address this issue, we further incorporate an unsupervised prior loss to substantially enhance training performance. Experimental results on benchmarks demonstrate that the proposed method, which requires neither clean data nor teacher models, performs on par with leading self-supervised and unsupervised approaches.

Recent studies in perceptual lossy compression have highlighted the advantage of stochastic coding with shared randomness between encoder and decoder, over deterministic encoding in the regime of “high perceptual quality”. While the theoretical benefits of stochastic coding have been well-established and demonstrated by analytic examples, its practical realization remains challenging and largely unexplored. In this work, we propose a practical learning framework for stochastic coding that effectively realizes its theoretical advantages. Starting with a theoretically optimal scheme, we develop an implementation closely approximates it through a two-stage training process: learning a stochastic encoder, followed by a stochastic decoder which is modeled as an optimal transport problem conditioned on minimum mean square error (MMSE) decoding. Additionally, for training stochastic coding models, we prove the equivalence between quantized representation and “noisy” representation. Based on this insight, we introduce a quantization-free training method that effectively addresses the non-differentiability challenge posed by quantization. Experiments on a circular distribution example and the MNIST dataset validate our findings and demonstrate the effectiveness of the proposed method.

Plane segmentation algorithms are widely used in robotics, serving key roles in scenarios such as indoor localization, scene understanding, and robotic manipulation. These applications typically require real-time, precise, and robust plane segmentation processing, which presents a significant challenge. Existing methods based on pixel-wise or fix-sized patch-wise operation are redundant, as planar regions in real-world scenes are of diverse sizes. In this paper, we introduce a highly efficient method for plane segmentation, namely Adaptive Patch-wise Region Growing (APRG). APRG begins with data sampling to construct a data pyramid. To avoid redundant planer fitting in large planar regions, we introduce an adaptive patch-wise plane fitting algorithm with the pyramid accessed in a top-down manner. The largest possible planar patches are obtained in this process. Subsequently we introduce a region growing algorithm specially designed for our patch representation. Overall, APRG achieves more than 600 FPS at a 640×480 resolution on a mid-range CPU without using parallel acceleration techniques, which outperforms the state-of-the-art method by a factor of 1.46. Besides, in addition to its speedup in run-time, APRG significantly improves the segmentation quality, especially on real-world data. Code is available at https://github.com/ZhangLanTao/APRG.git.