World models have demonstrated superiority in autonomous driving, particularly in the generation of multi-view driving videos. However, significant challenges still exist in generating customized driving videos. In this paper, we propose DriveDreamer-2, which incorporates a Large Language Model (LLM) to facilitate the creation of user-defined driving videos. Specifically, a trajectory generation function library is developed to produce trajectories that conform to user descriptions. Subsequently, an HDMap generator is designed to learn the mapping from trajectories to road structures. Ultimately, we propose the Unified Multi-View Model (UniMVM) to enhance temporal and spatial coherence in the generated multi-view driving videos. To the best of our knowledge, DriveDreamer-2 is the first world model to generate customized driving videos, and it can generate uncommon driving videos (e.g., vehicles abruptly cut in) in a user-friendly manner. Besides, experimental results demonstrate that the generated videos enhance the training of driving perception methods (e.g., 3D detection and tracking). Furthermore, video generation quality of DriveDreamer-2 surpasses other state-of-the-art methods, showcasing FID and FVD scores of 11.2 and 55.7, representing relative improvements of ~30% and ~50%.

Closed-loop simulation is crucial for end-to-end autonomous driving. Existing sensor simulation methods (e.g., NeRF and 3DGS) reconstruct driving scenes based on conditions that closely mirror training data distributions. However, these methods struggle with rendering novel trajectory, such as lane changes. Recent works have demonstrated that integrating world model knowledge alleviates these issues. Despite their efficiency, these approaches still encounter difficulties in the accurate representation of more complex maneuvers, with multi-lane shifts being a notable example. Therefore, we introduce ReconDreamer, which enhances driving scene reconstruction through incremental integration of world model knowledge. Specifically, DriveRestorer is proposed to mitigate artifacts via online restoration. This is complemented by a progressive data update strategy designed to ensure high-quality rendering for more complex maneuvers. To the best of our knowledge, ReconDreamer is the first method to effectively render in large maneuvers. Experimental results demonstrate that ReconDreamer outperforms Street Gaussians in the NTA-IoU, NTL-IoU, and FID, with relative improvements by 24.87%, 6.72%, and 29.97%. Furthermore, ReconDreamer surpasses DriveDreamer4D with PVG during large maneuver rendering, as verified by a relative improvement of 195.87% in the NTA-IoU metric and a user study.

Human-motion video generation has been a challenging task, primarily due to the difficulty inherent in learning human body movements. While some approaches have attempted to drive human-centric video generation explicitly through pose control, these methods typically rely on poses derived from existing videos, thereby lacking flexibility. To address this, we propose HumanDreamer, a decoupled human video generation framework that first generates diverse poses from text prompts and then leverages these poses to generate human-motion videos. Specifically, we propose MotionVid, the largest dataset for human-motion pose generation. Based on the dataset, we present MotionDiT, which is trained to generate structured human-motion poses from text prompts. Besides, a novel LAMA loss is introduced, which together contribute to a significant improvement in FID by 62.4%, along with respective enhancements in R-precision for top1, top2, and top3 by 41.8%, 26.3%, and 18.3%, thereby advancing both the Text-to-Pose control accuracy and FID metrics. Our experiments across various Pose-to-Video baselines demonstrate that the poses generated by our method can produce diverse and high-quality human-motion videos. Furthermore, our model can facilitate other downstream tasks, such as pose sequence prediction and 2D-3D motion lifting.

Closed-loop simulation is essential for advancing end-to-end autonomous driving systems. Contemporary sensor simulation methods, such as NeRF and 3DGS, rely predominantly on conditions closely aligned with training data distributions, which are largely confined to forward-driving scenarios. Consequently, these methods face limitations when rendering complex maneuvers (e.g., lane change, acceleration, deceleration). Recent advancements in autonomous-driving world models have demonstrated the potential to generate diverse driving videos. However, these approaches remain constrained to 2D video generation, inherently lacking the spatiotemporal coherence required to capture intricacies of dynamic driving environments. In this paper, we introduce DriveDreamer4D, which enhances 4D driving scene representation leveraging world model priors. Specifically, we utilize the world model as a data machine to synthesize novel trajectory videos, where structured conditions are explicitly leveraged to control the spatial-temporal consistency of traffic elements. Besides, the cousin data training strategy is proposed to facilitate merging real and synthetic data for optimizing 4DGS. To our knowledge, DriveDreamer4D is the first to utilize video generation models for improving 4D reconstruction in driving scenarios. Experimental results reveal that DriveDreamer4D significantly enhances generation quality under novel trajectory views, achieving a relative improvement in FID by 32.1%, 46.4%, and 16.3% compared to PVG, S3Gaussian, and Deformable-GS. Moreover, DriveDreamer4D markedly enhances the spatiotemporal coherence of driving agents, which is verified by a comprehensive user study and the relative increases of 22.6%, 43.5%, and 15.6% in the NTA-IoU metric.

BEV perception is of great importance in the field of autonomous driving, serving as the cornerstone of planning, controlling, and motion prediction. The quality of the BEV feature highly affects the performance of BEV perception. However, taking the noises in camera parameters and LiDAR scans into consideration, we usually obtain BEV representation with harmful noises. Diffusion models naturally have the ability to denoise noisy samples to the ideal data, which motivates us to utilize the diffusion model to get a better BEV representation. In this work, we propose an end-to-end framework, named DiffBEV, to exploit the potential of diffusion model to generate a more comprehensive BEV representation. To the best of our knowledge, we are the first to apply diffusion model to BEV perception. In practice, we design three types of conditions to guide the training of the diffusion model which denoises the coarse samples and refines the semantic feature in a progressive way. What's more, a cross-attention module is leveraged to fuse the context of BEV feature and the semantic content of conditional diffusion model. DiffBEV achieves a 25.9% mIoU on the nuScenes dataset, which is 6.2% higher than the best-performing existing approach. Quantitative and qualitative results on multiple benchmarks demonstrate the effectiveness of DiffBEV in BEV semantic segmentation and 3D object detection tasks.

Self-supervised monocular methods can efficiently learn depth information of weakly textured surfaces or reflective objects. However, the depth accuracy is limited due to the inherent ambiguity in monocular geometric modeling. In contrast, multi-frame depth estimation methods improve depth accuracy thanks to the success of Multi-View Stereo (MVS), which directly makes use of geometric constraints. Unfortunately, MVS often suffers from texture-less regions, non-Lambertian surfaces, and moving objects, especially in real-world video sequences without known camera motion and depth supervision. Therefore, we propose MOVEDepth, which exploits the MOnocular cues and VElocity guidance to improve multi-frame Depth learning. Unlike existing methods that enforce consistency between MVS depth and monocular depth, MOVEDepth boosts multi-frame depth learning by directly addressing the inherent problems of MVS. The key of our approach is to utilize monocular depth as a geometric priority to construct MVS cost volume, and adjust depth candidates of cost volume under the guidance of predicted camera velocity. We further fuse monocular depth and MVS depth by learning uncertainty in the cost volume, which results in a robust depth estimation against ambiguity in multi-view geometry. Extensive experiments show MOVEDepth achieves state-of-the-art performance: Compared with Monodepth2 and PackNet, our method relatively improves the depth accuracy by 20% and 19.8% on the KITTI benchmark. MOVEDepth also generalizes to the more challenging DDAD benchmark, relatively outperforming ManyDepth by 7.2%. The code is available at https://github.com/JeffWang987/MOVEDepth.

Recently, open-vocabulary learning has emerged to accomplish segmentation for arbitrary categories of text-based descriptions, which popularizes the segmentation system to more general-purpose application scenarios. However, existing methods devote to designing specialized architectures or parameters for specific segmentation tasks. These customized design paradigms lead to fragmentation between various segmentation tasks, thus hindering the uniformity of segmentation models. Hence in this paper, we propose FreeSeg, a generic framework to accomplish Unified, Universal and Open-Vocabulary Image Segmentation. FreeSeg optimizes an all-in-one network via one-shot training and employs the same architecture and parameters to handle diverse segmentation tasks seamlessly in the inference procedure. Additionally, adaptive prompt learning facilitates the unified model to capture task-aware and category-sensitive concepts, improving model robustness in multi-task and varied scenarios. Extensive experimental results demonstrate that FreeSeg establishes new state-of-the-art results in performance and generalization on three segmentation tasks, which outperforms the best task-specific architectures by a large margin: 5.5% mIoU on semantic segmentation, 17.6% mAP on instance segmentation, 20.1% PQ on panoptic segmentation for the unseen class on COCO. Project page: https://FreeSeg.github.io.

In recent years, vision-centric perception has flourished in various autonomous driving tasks, including 3D detection, semantic map construction, motion forecasting, and depth estimation. Nevertheless, the latency of vision-centric approaches is too high for practical deployment (e.g., most camera-based 3D detectors have a runtime greater than 300ms). To bridge the gap between ideal researches and real-world applications, it is necessary to quantify the trade-off between performance and efficiency. Traditionally, autonomous-driving perception benchmarks perform the offline evaluation, neglecting the inference time delay. To mitigate the problem, we propose the Autonomous-driving StreAming Perception (ASAP) benchmark, which is the first benchmark to evaluate the online performance of vision-centric perception in autonomous driving. On the basis of the 2Hz annotated nuScenes dataset, we first propose an annotation-extending pipeline to generate high-frame-rate labels for the 12Hz raw images. Referring to the practical deployment, the Streaming Perception Under constRained-computation (SPUR) evaluation protocol is further constructed, where the 12Hz inputs are utilized for streaming evaluation under the constraints of different computational resources. In the ASAP benchmark, comprehensive experiment results reveal that the model rank alters under different constraints, suggesting that the model latency and computation budget should be considered as design choices to optimize the practical deployment. To facilitate further research, we establish baselines for camera-based streaming 3D detection, which consistently enhance the streaming performance across various hardware. ASAP project page: https://github.com/JeffWang987/ASAP.

Semantic occupancy perception is essential for autonomous driving, as automated vehicles require a fine-grained perception of the 3D urban structures. However, existing relevant benchmarks lack diversity in urban scenes, and they only evaluate front-view predictions. Towards a comprehensive benchmarking of surrounding perception algorithms, we propose OpenOccupancy, which is the first surrounding semantic occupancy perception benchmark. In the OpenOccupancy benchmark, we extend the large-scale nuScenes dataset with dense semantic occupancy annotations. Previous annotations rely on LiDAR points superimposition, where some occupancy labels are missed due to sparse LiDAR channels. To mitigate the problem, we introduce the Augmenting And Purifying (AAP) pipeline to ~ 2× densify the annotations, where ∼4000 human hours are involved in the labeling process. Besides, camera-based, LiDAR-based and multi-modal baselines are established for the OpenOccupancy benchmark. Furthermore, considering the complexity of surrounding occupancy perception lies in the computational burden of high-resolution 3D predictions, we propose the Cascade Occupancy Network (CONet) to refine the coarse prediction, which relatively enhances the performance by ∼30% than the baseline. We hope the OpenOccupancy benchmark‡will boost the development of surrounding occupancy perception algorithms.

Restoring an accurate Bird's Eye View (BEV) map plays a crucial role in the perception of autonomous driving. The existing works of lifting representations from frontal view to BEV can be classified into two categories, i.e., Camera model-Based Feature Transformation (CBFT) and Camera model-Free Feature Transformation (CFFT). We empirically analyze the significant differences between CBFT and CFFT. The former method lift perspective features based on the flat- world assumption, which often causes distortion of regions lying above the ground plane. The latter method is limited in the perception performance due to the absence of geometric priors and time-consuming computing. In this paper, we propose a novel framework with a Hybrid Feature Transformation module (HFT) to lift perspective representations. Furthermore, we design a mutual learning scheme to augment hybrid transformation. The deformable attention mechanism enables the model to pay more attention to relevant regions and capture features with more semantics. We illustrate the effectiveness of HFT in BEV perception tasks, such as segmentation and object detection. Notably, in the task of semantic segmentation, extensive experiments demonstrate that HFT outperforms the previous state-of-the-art method by relatively 17.9% on the Argoverse and 22.0% on the KITTI 3D Object dataset. With negligible computing budget, HFT outperforms existing image- based methods on 3D object detection. The code will be released soon.

Neural architecture search (NAS) depends heavily on an efficient and accurate performance estimator. To speed up the evaluation process, recent advances, like differentiable architecture search (DARTS) and One-Shot approaches, instead of training every model from scratch, train a weight-sharing super-network to reuse parameters among different candidates, in which all child models can be efficiently evaluated. Though these methods significantly boost search efficiency, they inherently suffer from inaccurate and unstable performance estimation. To this end, we propose a general and effective framework for powering weight-sharing NAS, namely, PWSNAS, by shrinking search space automatically, i.e., candidate operators will be discarded if they are less important. With the strategy, our approach can provide a promising search space of a smaller size by progressively simplifying the original search space, which can reduce difficulties for existing NAS methods to find superior architectures. In particular, we present two strategies to guide the shrinking process: detect redundant operators with a new angle-based metric and decrease the degree of weight sharing of a super-network by increasing parameters, which differentiates PWSNAS from existing shrinking methods. Comprehensive analysis experiments on NASBench-201 verify the superiority of our proposed metric over existing accuracy-based and magnitude-based metrics. PWSNAS can easily apply to the state-of-the-art NAS methods, e.g., single path one-shot neural architecture search (SPOS), FairNAS, ProxylessNAS, DARTS, and progressive DARTS (PDARTS). We evaluate PWSNAS and demonstrate consistent performance gains over baseline methods.

Image-level weakly supervised semantic segmentation (WSSS) is a fundamental yet challenging computer vision task facilitating scene understanding and automatic driving. Most existing methods resort to classification-based Class Activation Maps (CAMs) to play as the initial pseudo labels, which tend to focus on the discriminative image regions and lack customized characteristics for the segmentation task. To alleviate this issue, we propose a novel activation modulation and recalibration (AMR) scheme, which leverages a spotlight branch and a compensation branch to obtain weighted CAMs that can provide recalibration supervision and task-specific concepts. Specifically, an attention modulation module (AMM) is employed to rearrange the distribution of feature importance from the channel-spatial sequential perspective, which helps to explicitly model channel-wise interdependencies and spatial encodings to adaptively modulate segmentation-oriented activation responses. Furthermore, we introduce a cross pseudo supervision for dual branches, which can be regarded as a semantic similar regularization to mutually refine two branches. Extensive experiments show that AMR establishes a new state-of-the-art performance on the PASCAL VOC 2012 dataset, surpassing not only current methods trained with the image-level of supervision but also some methods relying on stronger supervision, such as saliency label. Experiments also reveal that our scheme is plug-and-play and can be incorporated with other approaches to boost their performance. Our code is available at: https://github.com/jieqin-ai/AMR.

Most recent semisupervised video object segmentation (VOS) methods rely on fine-tuning deep convolutional neural networks online using the given mask of the first frame or predicted masks of subsequent frames. However, the online fine-tuning process is usually time-consuming, limiting the practical use of such methods. We propose a directional deep embedding and appearance learning (DDEAL) method, which is free of the online fine-tuning process, for fast VOS. First, a global directional matching module (GDMM), which can be efficiently implemented by parallel convolutional operations, is proposed to learn a semantic pixel-wise embedding as an internal guidance. Second, an effective directional appearance model-based statistics is proposed to represent the target and background on a spherical embedding space for VOS. Equipped with the GDMM and the directional appearance model learning module, DDEAL learns static cues from the labeled first frame and dynamically updates cues of the subsequent frames for object segmentation. Our method exhibits the state-of-the-art VOS performance without using online fine-tuning. Specifically, it achieves a J & F mean score of 74.8% on DAVIS 2017 data set and an overall score G of 71.3% on the large-scale YouTube-VOS data set, while retaining a speed of 25 fps with a single NVIDIA TITAN Xp GPU. Furthermore, our faster version runs 31 fps with only a little accuracy loss.

The tracking-by-detection tracking framework usually consists of two stages: drawing samples around the target object and classifying each sample as either the target object or background. Current popular trackers under this framework typically draw many samples from the raw image and feed them into the deep neural networks, resulting in high computational burden and low tracking speed. In this article, we propose an adversarial feature sampling learning (AFSL) method to address this problem. A convolutional neural network is designed, which takes only one cropped image around the target object as input, and samples are collected from the feature maps with spatial bilinear resampling. To enrich the appearance variations of positive samples in the feature space, which has limited spatial resolution, we fuse the high-level features and low-level features to better describe the target by using a generative adversarial network. Extensive experiments on benchmark data sets demonstrate that the proposed ASFL achieves leading tracking accuracy while significantly accelerating the speed of tracking-by-detection trackers.

This paper studies panoptic segmentation, a recently proposed task which segments foreground (FG) objects at the instance level as well as background (BG) contents at the semantic level. Existing methods mostly dealt with these two problems separately, but in this paper, we reveal the underlying relationship between them, in particular, FG objects provide complementary cues to assist BG understanding. Our approach, named the Attention-guided Unified Network (AUNet), is a unified framework with two branches for FG and BG segmentation simultaneously. Two sources of attentions are added to the BG branch, namely, RPN and FG segmentation mask to provide object-level and pixel-level attentions, respectively. Our approach is generalized to different backbones with consistent accuracy gain in both FG and BG segmentation, and also sets new state-of-the-arts both in the MS-COCO (46.5% PQ) and Cityscapes (59.0% PQ) benchmarks.

Deep neural networks (DNNs) have achieved great success in a wide range of computer vision areas, but the applications to mobile devices is limited due to their high storage and computational cost. Much efforts have been devoted to compress DNNs. In this paper, we propose a simple yet effective method for deep networks compression, named Cluster Regularized Quantization (CRQ), which can reduce the presentation precision of a full-precision model to ternary values without significant accuracy drop. In particular, the proposed method aims at reducing the quantization error by introducing a cluster regularization term, which is imposed on the full-precision weights to enable them naturally concentrate around the target values. Through explicitly regularizing the weights during the re-training stage, the full-precision model can achieve the smooth transition to the low-bit one. Comprehensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method.

Channel pruning, which seeks to reduce the model size by removing redundant channels, is a popular solution for deep networks compression. Existing channel pruning methods usually conduct layer-wise channel selection by directly minimizing the reconstruction error of feature maps between the baseline model and the pruned one. However, they ignore the feature and semantic distributions within feature maps and real contribution of channels to the overall performance. In this paper, we propose a new channel pruning method by explicitly using both intermediate outputs of the baseline model and the classification loss of the pruned model to supervise layer-wise channel selection. Particularly, we introduce an additional loss to encode the differences in the feature and semantic distributions within feature maps between the baseline model and the pruned one. By considering the reconstruction error, the additional loss and the classification loss at the same time, our approach can significantly improve the performance of the pruned model. Comprehensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method.

In this paper, a position measurement system, including drogue's landmark detection and position computation for autonomous aerial refueling of unmanned aerial vehicles, is proposed. A multitask parallel deep convolution neural network (MPDCNN) is designed to detect the landmarks of the drogue target. In MPDCNN, two parallel convolution networks are used, and a fusion mechanism is proposed to accomplish the effective fusion of the drogue's two salient parts' landmark detection. Considering the drogue target's geometric constraints, a position measurement method based on monocular vision is proposed. An effective fusion strategy, which fuses the measurement results of drogue's different parts, is proposed to achieve robust position measurement. The error of landmark detection with the proposed method is 3.9%, and it is obviously lower than the errors of other methods. Experimental results on the two KUKA robots platform verify the effectiveness and robustness of the proposed position measurement system for aerial refueling.

In recent years, the application of deep learning based on deep convolutional neural networks has gained great success in face detection. However, one of the remaining open challenges is the detection of small-scaled faces. The depth of the convolutional network can cause the projected feature map for small faces to be quickly shrunk, and most detection approaches with scale invariant can hardly handle less than 15×15 pixel faces. To solve this problem, we propose a different scales face detector (DSFD) based on Faster R-CNN. The new network can improve the precision of face detection while performing as real-time a Faster R-CNN. First, an efficient multitask region proposal network (RPN), combined with boosting face detection, is developed to obtain the human face ROI. Setting the ROI as a constraint, an anchor is inhomogeneously produced on the top feature map by the multitask RPN. A human face proposal is extracted through the anchor combined with facial landmarks. Then, a parallel-type Fast R-CNN network is proposed based on the proposal scale. According to the different percentages they cover on the images, the proposals are assigned to three corresponding Fast R-CNN networks. The three networks are separated through the proposal scales and differ from each other in the weight of feature map concatenation. A variety of strategies is introduced in our face detection network, including multitask learning, feature pyramid, and feature concatenation. Compared to state-of-the-art face detection methods such as UnitBox, HyperFace, FastCNN, the proposed DSFD method achieves promising performance on popular benchmarks including FDDB, AFW, PASCAL faces, and WIDER FACE.

Facial expression recognition is a challenging task in computer vision field using only single facial image. As we know, human faces are convex spheres. The self-occlusion phenomenon generated from face pose will seriously affect the accuracy of expression recognition. In order to solve this problem, we propose a novel facial expression recognition method for different pose faces based on special landmark detection (FER-MPI-SFL). Our method is based on two shared networks. The outputs of the first Network are 29 special landmarks and 1 face box, which are the inputs of the second network and used to estimate face pose. The methods of RoIAlign and feature map concatenation are introduced in the second network to recognize the facial expression. The weight allocation of feature maps concatenation is guided by the result of pose estimation. In addition, an improved center loss is proposed to make the distances between the features of different expressions larger and easier to be classified in the feature space. As a result, superior performance to other state-of-the-art methods is achieved in facial expression databases CK+, MMI, Oulu-CASIA VIS and a new created database CASIA-MFE which contains more faces with different poses.

In this paper, we propose a robust state-based structured support vector machine (SVM) tracking algorithm combined with incremental principal component analysis (PCA). Different from the current structured SVM for tracking, our method directly learns and predicts the object's states and not the 2-D translation transformation during tracking. We define the object's virtual state to combine the state-based structured SVM and incremental PCA. The virtual state is considered as the most confident state of the object in every frame. The incremental PCA is used to update the virtual feature vector corresponding to the virtual state and the principal subspace of the object's feature vectors. In order to improve the accuracy of the prediction, all the feature vectors are projected onto the principal subspace in the learning and prediction process of the state-based structured SVM. Experimental results on several challenging video sequences validate the effectiveness and robustness of our approach.