Zuotao Ning

High-definition (HD) maps play a crucial role in autonomous driving by providing a reliable foundation for behavior prediction and path planning. Recent methods model map elements as point sets and employ Transformer-based detection frameworks for end-to-end vectorized map construction. However, these approaches do not fully exploit geometric structural properties, thereby limiting the accuracy of predicting the shape and position of map elements. To address this limitation, we propose a novel method called GraphMapTR, which models map elements as graph structures. Each map element is represented as a subgraph where key points serve as nodes and edges capture structural continuity. To optimize the node query embeddings within subgraphs, we design a Graph Transformer-based decoder comprising three core components: Multi-head Self-Attention Between Subgraphs (MHSA-BSG), which facilitates efficient feature interaction across different subgraphs while enhancing node attention to their respective subgraph regions; Local Geometric-aware Self-Attention Within Subgraphs (LGSA-WSG), which dynamically aggregates global node relationships and local edge structures to enrich node representations; and Deformable Cross-Attention (DCA-BEV), which refines node embeddings through interaction with BEV features. In LGSA-WSG, priors from the rasterized map segmentation branch are incorporated to generate local geometric consistency scores (LGC-scores), which constitute edge structural bias. Extensive experiments demonstrate that GraphMapTR significantly outperforms state-of-the-art methods. In particular, it outperforms the state-of-the-art algorithm MapTRv2 by 5.0% mAP and 3.4% mAP on the nuScenes and Argoverse2 datasets, respectively.

The online vectorized map construction technique employs a neural network model to forecast the vectorized representation of a specific region around automobiles, using data obtained from sensors mounted on automobiles. Due to advances in end-to-end object detection with transformers framework, the research on query-based online mapping has attracted substantial attention. However, the fixed number of queries and the random initialization of query embeddings constrain the model's performance. Moreover, the transformer architecture for object detection is based on the assumption that queries are identically distributed and independent, a premise that is not entirely applicable to map point queries which possess established subordinate relationships with map element instances. To address these issues, we initially incorporate a simplified transformer layer that utilizes semantic priors in bird's-eye view features for query initialization. The queries are then sent to a transformer-based map decoder for optimization and combined with a dynamic query management mechanism to eliminate low-confidence queries, hence maintaining computational efficiency. Furthermore, to guarantee that point queries within each instance preserve a consistent representation and avoid feature confusion among map element instances, we proposed an instance internal consistency map decoder. We conduct extensive experiments on commonly used map construction datasets to evaluate the proposed method. The experimental results demonstrate that our proposed method achieves state-of-the-art performance on the nuScenes and Argoverse 2 datasets.

Visual Place Recognition (VPR) is crucial for various computer vision and robotics applications. Traditional VPR techniques relying on handcrafted features, have been enhanced by using Convolutional Neural Networks (CNNs). Recently, MixVPR has set new benchmarks in VPR by using advanced feature aggregation techniques. However, MixVPR's full-image feature mixing approach can lead to the ignoring of critical local detail information and regional saliency information in large-scale images. To overcome this, we propose MixVPR++, which integrates an Adaptive Gabor Texture Fuser with a Learnable Gabor Filter for enriching semantic context with texture details information and a Hierarchical-Region Feature-Mixer for better spatial hierarchy capture regional saliency information, thereby enhancing robustness and accuracy. Extensive experiments demonstrate that MixVPR++ outperforms state-of-the-art methods across most challenging benchmarks. Despite its impressive performance, MixVPR++ shows limitations in handling severe viewpoint changes, indicating an area for future improvement.

Visual Place Recognition (VPR) is a fundamental task in robotics and computer vision, where the ability to recognize locations from visual inputs is crucial for autonomous navigation systems. Traditional methods, which rely on handcrafted features or standard convolutional neural networks (CNNs), struggle with environmental changes that significantly alter a place’s appearance. Recent advancements in deep learning have improved VPR by focusing on deep-learned features, enhancing robustness under varying conditions. However, these methods often overlook saliency cues, leading to inefficiencies in dynamic scenes. To address these limitations, we propose an improved MixVPR model that incorporates both self-attention and cross-attention mechanisms through a spatial-wise hybrid attention mechanism. This enhancement integrates spatial saliency cues into the global image embedding, improving accuracy and reliability. We also utilize the DINOv2 visual transformer for robust feature extraction. Extensive experiments on mainstream VPR benchmarks demonstrate that our method achieves superior performance while maintaining computational efficiency. Ablation studies and visualizations further validate the contributions of our attention mechanisms to the model’s performance improvement.

The letter introduces a novel Visual Place Recognition (VPR) method called Lightweight Pooling-centric Saliency-aware VPR (LPS-VPR), a high-performance VPR method capable of exploiting saliency information without computational burden. The key contribution of the method is a pooling-based saliency encoder (PSE) that efficiently extracts and integrates local context saliency cues into the image embedding, avoiding using computationally intensive operations such as attention for saliency cues extraction. The method employs a multi-level feature pyramid fusion with a cascaded backtracking structure, merging multi-scale embedding information from different levels of the backbone network with negligible computational demand. The LPS-VPR method demonstrates superior performance on mainstream VPR benchmarks, showcasing state-of-the-art (SoTA) runtime efficiency and detailed ablation studies highlight the effectiveness of each LPS-VPR component. Extended visual demonstrations further underscore the effectiveness and superiority of LPS-VPR, making it a promising solution for real-world large-scale VPR applications.

Detection based on LiDAR and camera fusion is increasingly popular for researchers in the autonomous driving domain. Compared to the camera-only and LiDAR-only methods, the fusion-based methods indeed improve the detection accuracy on public-available datasets. However, due to the complexity of the projection or fusion mechanism, few of these methods can run in real time even on an advanced desktop GPU. Thus, in this paper, we propose a new fusion detection model G-Fusion with a light and fast image view-transform module. According to our receptive field analysis of image feature maps, we directly project image features to only one voxel layer located on the ground, then fuse the LiDAR and image features by concatenation and convolution. With this delicately designed module, G-Fusion greatly boosts the state-of-the-art speed performance on the nuScenes dataset, achieving a good balance with the competitive detection scores. Meanwhile, since the precision of sensor extrinsic parameters is important for most fusion-based methods, we also deeply dig into our model’s calibration error tolerance ability and discover the failure noise condition.