Deep learning has recently demonstrated its excellent performance on the task of multi-view stereo (MVS). However, loss functions applied for deep MVS are rarely studied. In this paper, we first analyze existing loss functions' properties for deep depth based MVS approaches. Regression based loss leads to inaccurate continuous results by computing mathematical expectation, while classification based loss outputs discretized depth values. To this end, we then propose a novel loss function, named adaptive Wasserstein loss, which is able to narrow down the difference between the true and predicted probability distributions of depth. Besides, a simple but effective offset module is introduced to better achieve sub-pixel prediction accuracy. Extensive experiments on different benchmarks, including DTU, Tanks and Temples and BlendedMVS, show that the proposed method with the adaptive Wasserstein loss and the offset module achieves state-of-the-art performance.
In this paper, we draw inspiration from Alberto Elfes' pioneering work in 1989, where he introduced the concept of the occupancy grid as World Models for robots. We imbue the robot with a spatial-temporal world model, termed UniWorld, to perceive its surroundings and predict the future behavior of other participants. UniWorld involves initially predicting 4D geometric occupancy as the World Models for foundational stage and subsequently fine-tuning on downstream tasks. UniWorld can estimate missing information concerning the world state and predict plausible future states of the world. Besides, UniWorld's pre-training process is label-free, enabling the utilization of massive amounts of image-LiDAR pairs to build a Foundational Model.The proposed unified pre-training framework demonstrates promising results in key tasks such as motion prediction, multi-camera 3D object detection, and surrounding semantic scene completion. When compared to monocular pre-training methods on the nuScenes dataset, UniWorld shows a significant improvement of about 1.5% in IoU for motion prediction, 2.0% in mAP and 2.0% in NDS for multi-camera 3D object detection, as well as a 3% increase in mIoU for surrounding semantic scene completion. By adopting our unified pre-training method, a 25% reduction in 3D training annotation costs can be achieved, offering significant practical value for the implementation of real-world autonomous driving. Codes are publicly available at https://github.com/chaytonmin/UniWorld.
Multi-camera 3D perception has emerged as a prominent research field in autonomous driving, offering a viable and cost-effective alternative to LiDAR-based solutions. However, existing multi-camera algorithms primarily rely on monocular image pre-training, which overlooks the spatial and temporal correlations among different camera views. To address this limitation, we propose the first multi-camera unified pre-training framework called Occ-BEV, which involves initially reconstructing the 3D scene as the foundational stage and subsequently fine-tuning the model on downstream tasks. Specifically, a 3D decoder is designed for leveraging Bird's Eye View (BEV) features from multi-view images to predict the 3D geometric occupancy to enable the model to capture a more comprehensive understanding of the 3D environment. A significant benefit of Occ-BEV is its capability of utilizing a considerable volume of unlabeled image-LiDAR pairs for pre-training purposes. The proposed multi-camera unified pre-training framework demonstrates promising results in key tasks such as multi-camera 3D object detection and surrounding semantic scene completion. When compared to monocular pre-training methods on the nuScenes dataset, Occ-BEV shows a significant improvement of about 2.0% in mAP and 2.0% in NDS for multi-camera 3D object detection, as well as a 3% increase in mIoU for surrounding semantic scene completion. Codes are publicly available at https://github.com/chaytonmin/Occ-BEV.
Learning accurate depth is essential to multi-view 3D object detection. Recent approaches mainly learn depth from monocular images, which confront inherent difficulties due to the ill-posed nature of monocular depth learning. Instead of using a sole monocular depth method, in this work, we propose a novel Surround-view Temporal Stereo (STS) technique that leverages the geometry correspondence between frames across time to facilitate accurate depth learning. Specifically, we regard the field of views from all cameras around the ego vehicle as a unified view, namely surroundview, and conduct temporal stereo matching on it. The resulting geometrical correspondence between different frames from STS is utilized and combined with the monocular depth to yield final depth prediction. Comprehensive experiments on nuScenes show that STS greatly boosts 3D detection ability, notably for medium and long distance objects. On BEVDepth with ResNet-50 backbone, STS improves mAP and NDS by 2.6% and 1.4%, respectively. Consistent improvements are observed when using a larger backbone and a larger image resolution, demonstrating its effectiveness
Mask-based pre-training has achieved great success for self-supervised learning in image, video, and language, without manually annotated supervision. However, it has not yet been studied about large-scale point clouds with redundant spatial information in autonomous driving. As the number of large-scale point clouds is huge, it is impossible to reconstruct the input point clouds. In this paper, we propose a mask voxel classification network for large-scale point clouds pre-training. Our key idea is to divide the point clouds into voxel representations and classify whether the voxel contains point clouds. This simple strategy makes the network to be voxel-aware of the object shape, thus improving the performance of the downstream tasks, such as 3D object detection. Our Voxel-MAE with even a 90% masking ratio can still learn representative features for the high spatial redundancy of large-scale point clouds. We also validate the effectiveness of Voxel-MAE in unsupervised domain adaptative tasks, which proves the generalization ability of Voxel-MAE. Our Voxel-MAE proves that it is feasible to pre-train large-scale point clouds without data annotations to enhance the perception ability of the autonomous vehicle. Extensive experiments show great effectiveness of our pre-trained model with 3D object detectors (SECOND, CenterPoint, and PV-RCNN) on two popular datasets (KITTI, Waymo). Codes are publicly available at https://github.com/chaytonmin/Voxel-MAE.
Freespace detection is an essential component of autonomous driving technology and plays an important role in trajectory planning. In the last decade, deep learning-based free space detection methods have been proved feasible. However, these efforts were focused on urban road environments and few deep learning-based methods were specifically designed for off-road free space detection due to the lack of off-road benchmarks. In this paper, we present the ORFD dataset, which, to our knowledge, is the first off-road free space detection dataset. The dataset was collected in different scenes (woodland, farmland, grassland, and countryside), different weather conditions (sunny, rainy, foggy, and snowy), and different light conditions (bright light, daylight, twilight, darkness), which totally contains 12,198 LiDAR point cloud and RGB image pairs with the traversable area, non-traversable area and unreachable area annotated in detail. We propose a novel network named OFF-Net, which unifies Transformer architecture to aggregate local and global information, to meet the requirement of large receptive fields for free space detection tasks. We also propose the cross-attention to dynamically fuse LiDAR and RGB image information for accurate off-road free space detection. Dataset and code are publicly available athttps://github.com/chaytonmin/OFF-Net.
In this paper, we present a novel recurrent multi-view stereo network based on long short-term memory (LSTM) with adaptive aggregation, namely AA-RMVSNet. We firstly introduce an intra-view aggregation module to adaptively extract image features by using context-aware convolution and multi-scale aggregation, which efficiently improves the performance on challenging regions, such as thin objects and large low-textured surfaces. To overcome the difficulty of varying occlusion in complex scenes, we propose an inter-view cost volume aggregation module for adaptive pixel-wise view aggregation, which is able to preserve better-matched pairs among all views. The two proposed adaptive aggregation modules are lightweight, effective and complementary regarding improving the accuracy and completeness of 3D reconstruction. Instead of conventional 3D CNNs, we utilize a hybrid network with recurrent structure for cost volume regularization, which allows high-resolution reconstruction and finer hypothetical plane sweep. The proposed network is trained end-to-end and achieves excellent performance on various datasets. It ranks $1^{st}$ among all submissions on Tanks and Temples benchmark and achieves competitive results on DTU dataset, which exhibits strong generalizability and robustness. Implementation of our method is available at https://github.com/QT-Zhu/AA-RMVSNet.
Deep learning has recently demonstrated its promising performance for vision-based parking-slot detection. However, very few existing methods explicitly take into account learning the link information of the marking-points, resulting in complex post-processing and erroneous detection. In this paper, we propose an attentional graph neural network based parking-slot detection method, which refers the marking-points in an around-view image as graph-structured data and utilize graph neural network to aggregate the neighboring information between marking-points. Without any manually designed post-processing, the proposed method is end-to-end trainable. Extensive experiments have been conducted on public benchmark dataset, where the proposed method achieves state-of-the-art accuracy. Code is publicly available at \url{https://github.com/Jiaolong/gcn-parking-slot}.