Underwater imaging is a critical task performed by marine robots for a wide range of applications including aquaculture, marine infrastructure inspection, and environmental monitoring. However, water column effects, such as attenuation and backscattering, drastically change the color and quality of imagery captured underwater. Due to varying water conditions and range-dependency of these effects, restoring underwater imagery is a challenging problem. This impacts downstream perception tasks including depth estimation and 3D reconstruction. In this paper, we advance state-of-the-art in neural radiance fields (NeRFs) to enable physics-informed dense depth estimation and color correction. Our proposed method, WaterNeRF, estimates parameters of a physics-based model for underwater image formation, leading to a hybrid data-driven and model-based solution. After determining the scene structure and radiance field, we can produce novel views of degraded as well as corrected underwater images, along with dense depth of the scene. We evaluate the proposed method qualitatively and quantitatively on a real underwater dataset.
Light-weight time-of-flight (ToF) depth sensors are small, cheap, low-energy and have been massively deployed on mobile devices for the purposes like autofocus, obstacle detection, etc. However, due to their specific measurements (depth distribution in a region instead of the depth value at a certain pixel) and extremely low resolution, they are insufficient for applications requiring high-fidelity depth such as 3D reconstruction. In this paper, we propose DELTAR, a novel method to empower light-weight ToF sensors with the capability of measuring high resolution and accurate depth by cooperating with a color image. As the core of DELTAR, a feature extractor customized for depth distribution and an attention-based neural architecture is proposed to fuse the information from the color and ToF domain efficiently. To evaluate our system in real-world scenarios, we design a data collection device and propose a new approach to calibrate the RGB camera and ToF sensor. Experiments show that our method produces more accurate depth than existing frameworks designed for depth completion and depth super-resolution and achieves on par performance with a commodity-level RGB-D sensor. Code and data are available at https://zju3dv.github.io/deltar/.
To achieve accurate 3D object detection at a low cost for autonomous driving, many multi-camera methods have been proposed and solved the occlusion problem of monocular approaches. However, due to the lack of accurate estimated depth, existing multi-camera methods often generate multiple bounding boxes along a ray of depth direction for difficult small objects such as pedestrians, resulting in an extremely low recall. Furthermore, directly applying depth prediction modules to existing multi-camera methods, generally composed of large network architectures, cannot meet the real-time requirements of self-driving applications. To address these issues, we propose Cross-view and Depth-guided Transformers for 3D Object Detection, CrossDTR. First, our lightweight depth predictor is designed to produce precise object-wise sparse depth maps and low-dimensional depth embeddings without extra depth datasets during supervision. Second, a cross-view depth-guided transformer is developed to fuse the depth embeddings as well as image features from cameras of different views and generate 3D bounding boxes. Extensive experiments demonstrated that our method hugely surpassed existing multi-camera methods by 10 percent in pedestrian detection and about 3 percent in overall mAP and NDS metrics. Also, computational analyses showed that our method is 5 times faster than prior approaches. Our codes will be made publicly available at https://github.com/sty61010/CrossDTR.
The perception system in personalized mobile agents requires developing indoor scene understanding models, which can understand 3D geometries, capture objectiveness, analyze human behaviors, etc. Nonetheless, this direction has not been well-explored in comparison with models for outdoor environments (e.g., the autonomous driving system that includes pedestrian prediction, car detection, traffic sign recognition, etc.). In this paper, we first discuss the main challenge: insufficient, or even no, labeled data for real-world indoor environments, and other challenges such as fusion between heterogeneous sources of information (e.g., RGB images and Lidar point clouds), modeling relationships between a diverse set of outputs (e.g., 3D object locations, depth estimation, and human poses), and computational efficiency. Then, we describe MMISM (Multi-modality input Multi-task output Indoor Scene understanding Model) to tackle the above challenges. MMISM considers RGB images as well as sparse Lidar points as inputs and 3D object detection, depth completion, human pose estimation, and semantic segmentation as output tasks. We show that MMISM performs on par or even better than single-task models; e.g., we improve the baseline 3D object detection results by 11.7% on the benchmark ARKitScenes dataset.
We introduce point affiliation into feature upsampling, a notion that describes the affiliation of each upsampled point to a semantic cluster formed by local decoder feature points with semantic similarity. By rethinking point affiliation, we present a generic formulation for generating upsampling kernels. The kernels encourage not only semantic smoothness but also boundary sharpness in the upsampled feature maps. Such properties are particularly useful for some dense prediction tasks such as semantic segmentation. The key idea of our formulation is to generate similarity-aware kernels by comparing the similarity between each encoder feature point and the spatially associated local region of decoder features. In this way, the encoder feature point can function as a cue to inform the semantic cluster of upsampled feature points. To embody the formulation, we further instantiate a lightweight upsampling operator, termed Similarity-Aware Point Affiliation (SAPA), and investigate its variants. SAPA invites consistent performance improvements on a number of dense prediction tasks, including semantic segmentation, object detection, depth estimation, and image matting. Code is available at: https://github.com/poppinace/sapa
In this paper, we present a fast monocular depth estimation method for enabling 3D perception capabilities of low-cost underwater robots. We formulate a novel end-to-end deep visual learning pipeline named UDepth, which incorporates domain knowledge of image formation characteristics of natural underwater scenes. First, we adapt a new input space from raw RGB image space by exploiting underwater light attenuation prior, and then devise a least-squared formulation for coarse pixel-wise depth prediction. Subsequently, we extend this into a domain projection loss that guides the end-to-end learning of UDepth on over 9K RGB-D training samples. UDepth is designed with a computationally light MobileNetV2 backbone and a Transformer-based optimizer for ensuring fast inference rates on embedded systems. By domain-aware design choices and through comprehensive experimental analyses, we demonstrate that it is possible to achieve state-of-the-art depth estimation performance while ensuring a small computational footprint. Specifically, with 70%-80% less network parameters than existing benchmarks, UDepth achieves comparable and often better depth estimation performance. While the full model offers over 66 FPS (13 FPS) inference rates on a single GPU (CPU core), our domain projection for coarse depth prediction runs at 51.5 FPS rates on single-board NVIDIA Jetson TX2s. The inference pipelines are available at https://github.com/uf-robopi/UDepth.
We propose DeepFusion, a modular multi-modal architecture to fuse lidars, cameras and radars in different combinations for 3D object detection. Specialized feature extractors take advantage of each modality and can be exchanged easily, making the approach simple and flexible. Extracted features are transformed into bird's-eye-view as a common representation for fusion. Spatial and semantic alignment is performed prior to fusing modalities in the feature space. Finally, a detection head exploits rich multi-modal features for improved 3D detection performance. Experimental results for lidar-camera, lidar-camera-radar and camera-radar fusion show the flexibility and effectiveness of our fusion approach. In the process, we study the largely unexplored task of faraway car detection up to 225 meters, showing the benefits of our lidar-camera fusion. Furthermore, we investigate the required density of lidar points for 3D object detection and illustrate implications at the example of robustness against adverse weather conditions. Moreover, ablation studies on our camera-radar fusion highlight the importance of accurate depth estimation.
A self-driving car must be able to reliably handle adverse weather conditions (e.g., snowy) to operate safely. In this paper, we investigate the idea of turning sensor inputs (i.e., images) captured in an adverse condition into a benign one (i.e., sunny), upon which the downstream tasks (e.g., semantic segmentation) can attain high accuracy. Prior work primarily formulates this as an unpaired image-to-image translation problem due to the lack of paired images captured under the exact same camera poses and semantic layouts. While perfectly-aligned images are not available, one can easily obtain coarsely-paired images. For instance, many people drive the same routes daily in both good and adverse weather; thus, images captured at close-by GPS locations can form a pair. Though data from repeated traversals are unlikely to capture the same foreground objects, we posit that they provide rich contextual information to supervise the image translation model. To this end, we propose a novel training objective leveraging coarsely-aligned image pairs. We show that our coarsely-aligned training scheme leads to a better image translation quality and improved downstream tasks, such as semantic segmentation, monocular depth estimation, and visual localization.
Bounded by the inherent ambiguity of depth perception, contemporary camera-based 3D object detection methods fall into the performance bottleneck. Intuitively, leveraging temporal multi-view stereo (MVS) technology is the natural knowledge for tackling this ambiguity. However, traditional attempts of MVS are flawed in two aspects when applying to 3D object detection scenes: 1) The affinity measurement among all views suffers expensive computation cost; 2) It is difficult to deal with outdoor scenarios where objects are often mobile. To this end, we introduce an effective temporal stereo method to dynamically select the scale of matching candidates, enable to significantly reduce computation overhead. Going one step further, we design an iterative algorithm to update more valuable candidates, making it adaptive to moving candidates. We instantiate our proposed method to multi-view 3D detector, namely BEVStereo. BEVStereo achieves the new state-of-the-art performance (i.e., 52.5% mAP and 61.0% NDS) on the camera-only track of nuScenes dataset. Meanwhile, extensive experiments reflect our method can deal with complex outdoor scenarios better than contemporary MVS approaches. Codes have been released at https://github.com/Megvii-BaseDetection/BEVStereo.
There have been attempts to detect 3D objects by fusion of stereo camera images and LiDAR sensor data or using LiDAR for pre-training and only monocular images for testing, but there have been less attempts to use only monocular image sequences due to low accuracy. In addition, when depth prediction using only monocular images, only scale-inconsistent depth can be predicted, which is the reason why researchers are reluctant to use monocular images alone. Therefore, we propose a method for predicting absolute depth and detecting 3D objects using only monocular image sequences by enabling end-to-end learning of detection networks and depth prediction networks. As a result, the proposed method surpasses other existing methods in performance on the KITTI 3D dataset. Even when monocular image and 3D LiDAR are used together during training in an attempt to improve performance, ours exhibit is the best performance compared to other methods using the same input. In addition, end-to-end learning not only improves depth prediction performance, but also enables absolute depth prediction, because our network utilizes the fact that the size of a 3D object such as a car is determined by the approximate size.
We propose SF2SE3, a novel approach to estimate scene dynamics in form of a segmentation into independently moving rigid objects and their SE(3)-motions. SF2SE3 operates on two consecutive stereo or RGB-D images. First, noisy scene flow is obtained by application of existing optical flow and depth estimation algorithms. SF2SE3 then iteratively (1) samples pixel sets to compute SE(3)-motion proposals, and (2) selects the best SE(3)-motion proposal with respect to a maximum coverage formulation. Finally, objects are formed by assigning pixels uniquely to the selected SE(3)-motions based on consistency with the input scene flow and spatial proximity. The main novelties are a more informed strategy for the sampling of motion proposals and a maximum coverage formulation for the proposal selection. We conduct evaluations on multiple datasets regarding application of SF2SE3 for scene flow estimation, object segmentation and visual odometry. SF2SE3 performs on par with the state of the art for scene flow estimation and is more accurate for segmentation and odometry.
Transparent objects are widely used in industrial automation and daily life. However, robust visual recognition and perception of transparent objects have always been a major challenge. Currently, most commercial-grade depth cameras are still not good at sensing the surfaces of transparent objects due to the refraction and reflection of light. In this work, we present a transformer-based transparent object depth estimation approach from a single RGB-D input. We observe that the global characteristics of the transformer make it easier to extract contextual information to perform depth estimation of transparent areas. In addition, to better enhance the fine-grained features, a feature fusion module (FFM) is designed to assist coherent prediction. Our empirical evidence demonstrates that our model delivers significant improvements in recent popular datasets, e.g., 25% gain on RMSE and 21% gain on REL compared to previous state-of-the-art convolutional-based counterparts in ClearGrasp dataset. Extensive results show that our transformer-based model enables better aggregation of the object's RGB and inaccurate depth information to obtain a better depth representation. Our code and the pre-trained model will be available at https://github.com/yuchendoudou/TODE.
For monocular depth estimation, acquiring ground truths for real data is not easy, and thus domain adaptation methods are commonly adopted using the supervised synthetic data. However, this may still incur a large domain gap due to the lack of supervision from the real data. In this paper, we develop a domain adaptation framework via generating reliable pseudo ground truths of depth from real data to provide direct supervisions. Specifically, we propose two mechanisms for pseudo-labeling: 1) 2D-based pseudo-labels via measuring the consistency of depth predictions when images are with the same content but different styles; 2) 3D-aware pseudo-labels via a point cloud completion network that learns to complete the depth values in the 3D space, thus providing more structural information in a scene to refine and generate more reliable pseudo-labels. In experiments, we show that our pseudo-labeling methods improve depth estimation in various settings, including the usage of stereo pairs during training. Furthermore, the proposed method performs favorably against several state-of-the-art unsupervised domain adaptation approaches in real-world datasets.
Remarkable progress has been made in self-supervised monocular depth estimation (SS-MDE) by exploring cross-view consistency, e.g., photometric consistency and 3D point cloud consistency. However, they are very vulnerable to illumination variance, occlusions, texture-less regions, as well as moving objects, making them not robust enough to deal with various scenes. To address this challenge, we study two kinds of robust cross-view consistency in this paper. Firstly, the spatial offset field between adjacent frames is obtained by reconstructing the reference frame from its neighbors via deformable alignment, which is used to align the temporal depth features via a Depth Feature Alignment (DFA) loss. Secondly, the 3D point clouds of each reference frame and its nearby frames are calculated and transformed into voxel space, where the point density in each voxel is calculated and aligned via a Voxel Density Alignment (VDA) loss. In this way, we exploit the temporal coherence in both depth feature space and 3D voxel space for SS-MDE, shifting the "point-to-point" alignment paradigm to the "region-to-region" one. Compared with the photometric consistency loss as well as the rigid point cloud alignment loss, the proposed DFA and VDA losses are more robust owing to the strong representation power of deep features as well as the high tolerance of voxel density to the aforementioned challenges. Experimental results on several outdoor benchmarks show that our method outperforms current state-of-the-art techniques. Extensive ablation study and analysis validate the effectiveness of the proposed losses, especially in challenging scenes. The code and models are available at https://github.com/sunnyHelen/RCVC-depth.
Self-supervised monocular depth estimation has received much attention recently in computer vision. Most of the existing works in literature aggregate multi-scale features for depth prediction via either straightforward concatenation or element-wise addition, however, such feature aggregation operations generally neglect the contextual consistency between multi-scale features. Addressing this problem, we propose the Self-Distilled Feature Aggregation (SDFA) module for simultaneously aggregating a pair of low-scale and high-scale features and maintaining their contextual consistency. The SDFA employs three branches to learn three feature offset maps respectively: one offset map for refining the input low-scale feature and the other two for refining the input high-scale feature under a designed self-distillation manner. Then, we propose an SDFA-based network for self-supervised monocular depth estimation, and design a self-distilled training strategy to train the proposed network with the SDFA module. Experimental results on the KITTI dataset demonstrate that the proposed method outperforms the comparative state-of-the-art methods in most cases. The code is available at https://github.com/ZM-Zhou/SDFA-Net_pytorch.
Recent deep learning approaches for multi-view depth estimation are employed either in a depth-from-video or a multi-view stereo setting. Despite different settings, these approaches are technically similar: they correlate multiple source views with a keyview to estimate a depth map for the keyview. In this work, we introduce the Robust Multi-View Depth Benchmark that is built upon a set of public datasets and allows evaluation in both settings on data from different domains. We evaluate recent approaches and find imbalanced performances across domains. Further, we consider a third setting, where camera poses are available and the objective is to estimate the corresponding depth maps with their correct scale. We show that recent approaches do not generalize across datasets in this setting. This is because their cost volume output runs out of distribution. To resolve this, we present the Robust MVD Baseline model for multi-view depth estimation, which is built upon existing components but employs a novel scale augmentation procedure. It can be applied for robust multi-view depth estimation, independent of the target data. We provide code for the proposed benchmark and baseline model at https://github.com/lmb-freiburg/robustmvd.
Self-supervised depth learning from monocular images normally relies on the 2D pixel-wise photometric relation between temporally adjacent image frames. However, they neither fully exploit the 3D point-wise geometric correspondences, nor effectively tackle the ambiguities in the photometric warping caused by occlusions or illumination inconsistency. To address these problems, this work proposes Density Volume Construction Network (DevNet), a novel self-supervised monocular depth learning framework, that can consider 3D spatial information, and exploit stronger geometric constraints among adjacent camera frustums. Instead of directly regressing the pixel value from a single image, our DevNet divides the camera frustum into multiple parallel planes and predicts the pointwise occlusion probability density on each plane. The final depth map is generated by integrating the density along corresponding rays. During the training process, novel regularization strategies and loss functions are introduced to mitigate photometric ambiguities and overfitting. Without obviously enlarging model parameters size or running time, DevNet outperforms several representative baselines on both the KITTI-2015 outdoor dataset and NYU-V2 indoor dataset. In particular, the root-mean-square-deviation is reduced by around 4% with DevNet on both KITTI-2015 and NYU-V2 in the task of depth estimation. Code is available at https://github.com/gitkaichenzhou/DevNet.
Neural Radiance Fields (NeRF) achieve photo-realistic view synthesis with densely captured input images. However, the geometry of NeRF is extremely under-constrained given sparse views, resulting in significant degradation of novel view synthesis quality. Inspired by self-supervised depth estimation methods, we propose StructNeRF, a solution to novel view synthesis for indoor scenes with sparse inputs. StructNeRF leverages the structural hints naturally embedded in multi-view inputs to handle the unconstrained geometry issue in NeRF. Specifically, it tackles the texture and non-texture regions respectively: a patch-based multi-view consistent photometric loss is proposed to constrain the geometry of textured regions; for non-textured ones, we explicitly restrict them to be 3D consistent planes. Through the dense self-supervised depth constraints, our method improves both the geometry and the view synthesis performance of NeRF without any additional training on external data. Extensive experiments on several real-world datasets demonstrate that StructNeRF surpasses state-of-the-art methods for indoor scenes with sparse inputs both quantitatively and qualitatively.
Due to the rise of spherical cameras, monocular 360 depth estimation becomes an important technique for many applications (e.g., autonomous systems). Thus, state-of-the-art frameworks for monocular 360 depth estimation such as bi-projection fusion in BiFuse are proposed. To train such a framework, a large number of panoramas along with the corresponding depth ground truths captured by laser sensors are required, which highly increases the cost of data collection. Moreover, since such a data collection procedure is time-consuming, the scalability of extending these methods to different scenes becomes a challenge. To this end, self-training a network for monocular depth estimation from 360 videos is one way to alleviate this issue. However, there are no existing frameworks that incorporate bi-projection fusion into the self-training scheme, which highly limits the self-supervised performance since bi-projection fusion can leverage information from different projection types. In this paper, we propose BiFuse++ to explore the combination of bi-projection fusion and the self-training scenario. To be specific, we propose a new fusion module and Contrast-Aware Photometric Loss to improve the performance of BiFuse and increase the stability of self-training on real-world videos. We conduct both supervised and self-supervised experiments on benchmark datasets and achieve state-of-the-art performance.
Extracting depth information from a single RGB image is a fundamental and challenging task in computer vision with wide-ranging applications. This task cannot be solved using traditional methods like multi-view geometry but only by deep learning. Existing methods using convolutional neural nets produce inconsistent and blurry results due to the lack of long-range dependencies. With the recent success of Transformer networks in computer vision, which can process information locally and globally, we leverage this idea to propose a novel architecture named Focal-WNet in this paper. This architecture consists of two separate encoders and a single decoder. The main aim of this network is to learn most monocular depth cues like relative scale, contrast differences, texture gradient, etc. We incorporate focal self-attention instead of vanilla self-attention to reduce the computational complexity of the network. Along with the focal transformer layers, we leverage a convolutional architecture to learn depth cues that cannot be learned by a transformer alone as some cues like occlusion require a local receptive field and are easier for a conv-net to learn. Extensive experiments show that the proposed Focal-WNet achieves competitive results on two challenging datasets.