Shiqiang Hu

Camera–LiDAR fusion has shown great promise in advancing road detection for autonomous driving, combining the semantic richness of RGB imagery with the depth accuracy of LiDAR. However, its practical deployment faces two main challenges: (1) the high cost and limited accessibility of LiDAR sensors hinder their large-scale adoption, and (2) the scarcity of labeled data in unseen target domains limits the generalization capability of supervised methods. To overcome these limitations, we propose SPADA-Road, an unsupervised framework that integrates Superpixel-guided segmentation with a novel Pseudo-LiDAR (PL) generation module. Our PL module synthesizes depth cues from monocular RGB inputs, effectively augmenting training data without requiring physical LiDAR sensors. To enable robust cross-domain generalization, we adopt an adversarial domain adaptation strategy that aligns feature distributions between labeled source and unlabeled target domains. We evaluate SPADA-Road on the KITTI Road dataset for training, and validate its performance on two LiDAR-free benchmarks—Cityscapes and CamVid. Extensive experiments demonstrate that our method achieves superior performance compared to several state-of-the-art baselines, highlighting its effectiveness in LiDAR-free and label-scarce scenarios.

Maximum Mean Discrepancy (MMD) is widely used in domain adaptation (DA), yet standard MMD alignment is largely decision-boundary agnostic: reducing cross-domain distributional shift alone does not guarantee a tight target-risk bound. To address this limitation, we propose Decision Boundary-Aware MMD (DB-MMD), a boundary-aware discrepancy that augments MMD with two sample-dependent graphs, namely the Compacting Graph and the Separation Graph, to emphasize boundary-critical samples and enable joint optimization of distribution alignment and decision boundaries. The resulting kernelizable formulation integrates as a plug-in without adversarial training or architectural changes, and admits an efficient generalized-eigenvalue solution. When instantiated in four representative frameworks (JDA, CDDA, DGA-DA, MEDA), DB-MMD yields consistent improvements across eight benchmarks and 49 transfer tasks (up to +9.5 on PIE), achieving competitive or state-of-the-art performance against 50 methods. These results demonstrate that explicitly incorporating boundary awareness into discrepancy learning is an effective and broadly applicable principle for domain adaptation.

Domain Adaptive Semantic Segmentation (DASS) seeks to bridge the distribution gap between labeled source data and unlabeled target data for pixel-wise prediction. Recent generative approaches address this challenge by translating source images into the target domain, but most existing methods require training or fine-tuning a dedicated generator for each source-target pair, leading to high computational cost and limited scalability. In this work, we propose DIG-Seg, a re-training-free generative DASS framework that leverages a single diffusion model pre-trained on the target domain to generate semantic-consistent pseudo-target images from arbitrary source inputs via Diffusion Inverse Guidance (DIG). Specifically, DIG-Seg formulates DA as an inverse sampling process guided by a domain-invariant Fourier high-pass operator, which preserves semantic structures while adapting target-domain appearance. To avoid the sensitivity of diffusion inversion to manually tuned hyperparameters and to balance semantic fidelity with style realism, we further introduce a Metropolis-Hastings Hamiltonian Monte Carlo (MH-HMC) strategy that provides an interpretable and principled stopping criterion by selectively retaining high-confidence samples. The proposed framework is plug-and-play and can be seamlessly integrated into standard DASS pipelines without any per-pair generative re-training. Extensive experiments on GTA5→Cityscapes and SYNTHIA→Cityscapes with various segmentation backbones demonstrate that DIG-Seg consistently improves adaptation performance, offering an efficient and scalable solution for generative DASS.

Pseudo-labeling is a cornerstone of Unsupervised Domain Adaptation (UDA), yet the scarcity of High-Confidence Pseudo-Labeled Target Domain Samples (hcpl-tds) often leads to inaccurate cross-domain statistical alignment, causing DA failures. To address this challenge, we propose Noise Optimized Conditional Diffusion for Domain Adaptation (NOCDDA), which seamlessly integrates the generative capabilities of conditional diffusion models with the decision-making requirements of DA to achieve task-coupled optimization for efficient adaptation. For robust cross-domain consistency, we modify the DA classifier to align with the conditional diffusion classifier within a unified optimization framework, enabling forward training on noise-varying cross-domain samples. Furthermore, we argue that the conventional N(0,I) initialization in diffusion models often generates class-confused hcpl-tds, compromising discriminative DA. To resolve this, we introduce a class-aware noise optimization strategy that refines sampling regions for reverse class-specific hcpl-tds generation, effectively enhancing cross-domain alignment. Extensive experiments across 5 benchmark datasets and 29 DA tasks demonstrate significant performance gains of NOCDDA over 31 state-of-the-art methods, validating its robustness and effectiveness.

Camera-LiDAR fusion has shown great promise in advancing road detection for autonomous driving, combining the semantic richness of RGB imagery with the depth accuracy of LiDAR. However, its practical deployment faces two main challenges: (1) the high cost and limited accessibility of LiDAR sensors hinder their large-scale adoption, and (2) the scarcity of labeled data in unseen target domains limits the generalization capability of supervised methods. To overcome these limitations, we propose SPADA-Road, an unsupervised framework that integrates Superpixel-guided segmentation with a novel Pseudo-LiDAR (PL) generation module. Our PL module synthesizes depth cues from monocular RGB inputs, effectively augmenting training data without requiring physical LiDAR sensors. To enable robust cross-domain generalization, we adopt an adversarial domain adaptation strategy that aligns feature distributions between labeled source and unlabeled target domains. We evaluate SPADA-Road on the KITTI Road dataset for training, and validate its performance on two LiDAR-free benchmarks—Cityscapes and CamVid. Extensive experiments demonstrate that our method achieves superior performance compared to several state-of-the-art baselines, highlighting its effectiveness in LiDAR-free and label-scarce scenarios.

In domain adaptation (DA), the effectiveness of deep learning-based models is often constrained by batch learning strategies that fail to fully apprehend the global statistical and geometric characteristics of data distributions. Addressing this gap, we introduce “Global Awareness Enhanced Domain Adaptation” (GAN-DA), a novel approach that transcends traditional batch-based limitations. GAN-DA integrates a unique predefined feature representation (PFR) to facilitate the alignment of cross-domain distributions, thereby achieving a comprehensive global statistical awareness. This representation is innovatively expanded to encompass orthogonal and common feature aspects, which enhances the unification of global manifold structures and refines decision boundaries for more effective DA. Our extensive experiments, encompassing 27 diverse cross-domain image classification tasks, demonstrate GAN-DA's remarkable superiority, outperforming 24 established DA methods by a significant margin. Furthermore, our in-depth analyses shed light on the decision-making processes, revealing insights into the adaptability and efficiency of GAN-DA. This approach not only addresses the limitations of existing DA methodologies but also sets a new benchmark in the realm of domain adaptation, offering broad implications for future research and applications in this field.