Self-supervised learning is the key to unlocking generic computer vision systems. By eliminating the reliance on ground-truth annotations, it allows scaling to much larger data quantities. Unfortunately, self-supervised monocular depth estimation (SS-MDE) has been limited by the absence of diverse training data. Existing datasets have focused exclusively on urban driving in densely populated cities, resulting in models that fail to generalize beyond this domain. To address these limitations, this paper proposes two novel datasets: SlowTV and CribsTV. These are large-scale datasets curated from publicly available YouTube videos, containing a total of 2M training frames. They offer an incredibly diverse set of environments, ranging from snowy forests to coastal roads, luxury mansions and even underwater coral reefs. We leverage these datasets to tackle the challenging task of zero-shot generalization, outperforming every existing SS-MDE approach and even some state-of-the-art supervised methods. The generalization capabilities of our models are further enhanced by a range of components and contributions: 1) learning the camera intrinsics, 2) a stronger augmentation regime targeting aspect ratio changes, 3) support frame randomization, 4) flexible motion estimation, 5) a modern transformer-based architecture. We demonstrate the effectiveness of each component in extensive ablation experiments. To facilitate the development of future research, we make the datasets, code and pretrained models available to the public at https://github.com/jspenmar/slowtv_monodepth.
Given 2D point correspondences between an image pair, inferring the camera motion is a fundamental issue in the computer vision community. The existing works generally set out from the epipolar constraint and estimate the essential matrix, which is not optimal in the maximum likelihood (ML) sense. In this paper, we dive into the original measurement model with respect to the rotation matrix and normalized translation vector and formulate the ML problem. We then propose a two-step algorithm to solve it: In the first step, we estimate the variance of measurement noises and devise a consistent estimator based on bias elimination; In the second step, we execute a one-step Gauss-Newton iteration on manifold to refine the consistent estimate. We prove that the proposed estimate owns the same asymptotic statistical properties as the ML estimate: The first is consistency, i.e., the estimate converges to the ground truth as the point number increases; The second is asymptotic efficiency, i.e., the mean squared error of the estimate converges to the theoretical lower bound -- Cramer-Rao bound. In addition, we show that our algorithm has linear time complexity. These appealing characteristics endow our estimator with a great advantage in the case of dense point correspondences. Experiments on both synthetic data and real images demonstrate that when the point number reaches the order of hundreds, our estimator outperforms the state-of-the-art ones in terms of estimation accuracy and CPU time.
The estimation of implicit cross-frame correspondences and the high computational cost have long been major challenges in video semantic segmentation (VSS) for driving scenes. Prior works utilize keyframes, feature propagation, or cross-frame attention to address these issues. By contrast, we are the first to harness vanishing point (VP) priors for more effective segmentation. Intuitively, objects near VPs (i.e., away from the vehicle) are less discernible. Moreover, they tend to move radially away from the VP over time in the usual case of a forward-facing camera, a straight road, and linear forward motion of the vehicle. Our novel, efficient network for VSS, named VPSeg, incorporates two modules that utilize exactly this pair of static and dynamic VP priors: sparse-to-dense feature mining (DenseVP) and VP-guided motion fusion (MotionVP). MotionVP employs VP-guided motion estimation to establish explicit correspondences across frames and help attend to the most relevant features from neighboring frames, while DenseVP enhances weak dynamic features in distant regions around VPs. These modules operate within a context-detail framework, which separates contextual features from high-resolution local features at different input resolutions to reduce computational costs. Contextual and local features are integrated through contextualized motion attention (CMA) for the final prediction. Extensive experiments on two popular driving segmentation benchmarks, Cityscapes and ACDC, demonstrate that VPSeg outperforms previous SOTA methods, with only modest computational overhead.
Despite the promise of superior performance under challenging conditions, event-based motion estimation remains a hard problem owing to the difficulty of extracting and tracking stable features from event streams. In order to robustify the estimation, it is generally believed that fusion with other sensors is a requirement. In this work, we demonstrate reliable, purely event-based visual odometry on planar ground vehicles by employing the constrained non-holonomic motion model of Ackermann steering platforms. We extend single feature n-linearities for regular frame-based cameras to the case of quasi time-continuous event-tracks, and achieve a polynomial form via variable degree Taylor expansions. Robust averaging over multiple event tracks is simply achieved via histogram voting. As demonstrated on both simulated and real data, our algorithm achieves accurate and robust estimates of the vehicle's instantaneous rotational velocity, and thus results that are comparable to the delta rotations obtained by frame-based sensors under normal conditions. We furthermore significantly outperform the more traditional alternatives in challenging illumination scenarios. The code is available at \url{https://github.com/gowanting/NHEVO}.
Cone-beam computed tomography (CBCT) systems, with their flexibility, present a promising avenue for direct point-of-care medical imaging, particularly in critical scenarios such as acute stroke assessment. However, the integration of CBCT into clinical workflows faces challenges, primarily linked to long scan duration resulting in patient motion during scanning and leading to image quality degradation in the reconstructed volumes. This paper introduces a novel approach to CBCT motion estimation using a gradient-based optimization algorithm, which leverages generalized derivatives of the backprojection operator for cone-beam CT geometries. Building on that, a fully differentiable target function is formulated which grades the quality of the current motion estimate in reconstruction space. We drastically accelerate motion estimation yielding a 19-fold speed-up compared to existing methods. Additionally, we investigate the architecture of networks used for quality metric regression and propose predicting voxel-wise quality maps, favoring autoencoder-like architectures over contracting ones. This modification improves gradient flow, leading to more accurate motion estimation. The presented method is evaluated through realistic experiments on head anatomy. It achieves a reduction in reprojection error from an initial average of 3mm to 0.61mm after motion compensation and consistently demonstrates superior performance compared to existing approaches. The analytic Jacobian for the backprojection operation, which is at the core of the proposed method, is made publicly available. In summary, this paper contributes to the advancement of CBCT integration into clinical workflows by proposing a robust motion estimation approach that enhances efficiency and accuracy, addressing critical challenges in time-sensitive scenarios.
This paper overviews different pose representations and metric functions in visual odometry (VO) networks. The performance of VO networks heavily relies on how their architecture encodes the information. The choice of pose representation and loss function significantly impacts network convergence and generalization. We investigate these factors in the VO network DeepVO by implementing loss functions based on Euler, quaternion, and chordal distance and analyzing their influence on performance. The results of this study provide insights into how loss functions affect the designing of efficient and accurate VO networks for camera motion estimation. The experiments illustrate that a distance that complies with the mathematical requirements of a metric, such as the chordal distance, provides better generalization and faster convergence. The code for the experiments can be found at https://github.com/remaro-network/Loss_VO_right
Given the difficulty of manually annotating motion in video, the current best motion estimation methods are trained with synthetic data, and therefore struggle somewhat due to a train/test gap. Self-supervised methods hold the promise of training directly on real video, but typically perform worse. These include methods trained with warp error (i.e., color constancy) combined with smoothness terms, and methods that encourage cycle-consistency in the estimates (i.e., tracking backwards should yield the opposite trajectory as tracking forwards). In this work, we take on the challenge of improving state-of-the-art supervised models with self-supervised training. We find that when the initialization is supervised weights, most existing self-supervision techniques actually make performance worse instead of better, which suggests that the benefit of seeing the new data is overshadowed by the noise in the training signal. Focusing on obtaining a "clean" training signal from real-world unlabelled video, we propose to separate label-making and training into two distinct stages. In the first stage, we use the pre-trained model to estimate motion in a video, and then select the subset of motion estimates which we can verify with cycle-consistency. This produces a sparse but accurate pseudo-labelling of the video. In the second stage, we fine-tune the model to reproduce these outputs, while also applying augmentations on the input. We complement this boot-strapping method with simple techniques that densify and re-balance the pseudo-labels, ensuring that we do not merely train on "easy" tracks. We show that our method yields reliable gains over fully-supervised methods in real videos, for both short-term (flow-based) and long-range (multi-frame) pixel tracking.
The typical point cloud sampling methods used in state estimation for mobile robots preserve a high level of point redundancy. This redundancy unnecessarily slows down the estimation pipeline and may cause drift under real-time constraints. Such undue latency becomes a bottleneck for resource-constrained robots (especially UAVs), requiring minimal delay for agile and accurate operation. We propose a novel, deterministic, uninformed, and single-parameter point cloud sampling method named RMS that minimizes redundancy within a 3D point cloud. In contrast to the state of the art, RMS balances the translation-space observability by leveraging the fact that linear and planar surfaces inherently exhibit high redundancy propagated into iterative estimation pipelines. We define the concept of gradient flow, quantifying the local surface underlying a point. We also show that maximizing the entropy of the gradient flow minimizes point redundancy for robot ego-motion estimation. We integrate RMS into the point-based KISS-ICP and feature-based LOAM odometry pipelines and evaluate experimentally on KITTI, Hilti-Oxford, and custom datasets from multirotor UAVs. The experiments demonstrate that RMS outperforms state-of-the-art methods in speed, compression, and accuracy in well-conditioned as well as in geometrically-degenerated settings.
In magnetic resonance imaging (MRI), slice-to-volume reconstruction (SVR) refers to computational reconstruction of an unknown 3D magnetic resonance volume from stacks of 2D slices corrupted by motion. While promising, current SVR methods require multiple slice stacks for accurate 3D reconstruction, leading to long scans and limiting their use in time-sensitive applications such as fetal fMRI. Here, we propose a SVR method that overcomes the shortcomings of previous work and produces state-of-the-art reconstructions in the presence of extreme inter-slice motion. Inspired by the recent success of single-view depth estimation methods, we formulate SVR as a single-stack motion estimation task and train a fully convolutional network to predict a motion stack for a given slice stack, producing a 3D reconstruction as a byproduct of the predicted motion. Extensive experiments on the SVR of adult and fetal brains demonstrate that our fully convolutional method is twice as accurate as previous SVR methods. Our code is available at github.com/seannz/svr.
We present GaussianAvatar, an efficient approach to creating realistic human avatars with dynamic 3D appearances from a single video. We start by introducing animatable 3D Gaussians to explicitly represent humans in various poses and clothing styles. Such an explicit and animatable representation can fuse 3D appearances more efficiently and consistently from 2D observations. Our representation is further augmented with dynamic properties to support pose-dependent appearance modeling, where a dynamic appearance network along with an optimizable feature tensor is designed to learn the motion-to-appearance mapping. Moreover, by leveraging the differentiable motion condition, our method enables a joint optimization of motions and appearances during avatar modeling, which helps to tackle the long-standing issue of inaccurate motion estimation in monocular settings. The efficacy of GaussianAvatar is validated on both the public dataset and our collected dataset, demonstrating its superior performances in terms of appearance quality and rendering efficiency.
A truly generalizable approach to rigid segmentation and motion estimation is fundamental to 3D understanding of articulated objects and moving scenes. In view of the closely intertwined relationship between segmentation and motion estimates, we present an SE(3) equivariant architecture and a training strategy to tackle this task in an unsupervised manner. Our architecture is composed of two interconnected, lightweight heads. These heads predict segmentation masks using point-level invariant features and estimate motion from SE(3) equivariant features, all without the need for category information. Our training strategy is unified and can be implemented online, which jointly optimizes the predicted segmentation and motion by leveraging the interrelationships among scene flow, segmentation mask, and rigid transformations. We conduct experiments on four datasets to demonstrate the superiority of our method. The results show that our method excels in both model performance and computational efficiency, with only 0.25M parameters and 0.92G FLOPs. To the best of our knowledge, this is the first work designed for category-agnostic part-level SE(3) equivariance in dynamic point clouds.
In this paper, we propose an approach to address the problem of 3D reconstruction of scenes from a single image captured by a light-field camera equipped with a rolling shutter sensor. Our method leverages the 3D information cues present in the light-field and the motion information provided by the rolling shutter effect. We present a generic model for the imaging process of this sensor and a two-stage algorithm that minimizes the re-projection error while considering the position and motion of the camera in a motion-shape bundle adjustment estimation strategy. Thereby, we provide an instantaneous 3D shape-and-pose-and-velocity sensing paradigm. To the best of our knowledge, this is the first study to leverage this type of sensor for this purpose. We also present a new benchmark dataset composed of different light-fields showing rolling shutter effects, which can be used as a common base to improve the evaluation and tracking the progress in the field. We demonstrate the effectiveness and advantages of our approach through several experiments conducted for different scenes and types of motions. The source code and dataset are publicly available at: https://github.com/ICB-Vision-AI/RSLF
In this article, we investigate self-supervised 3D scene flow estimation and class-agnostic motion prediction on point clouds. A realistic scene can be well modeled as a collection of rigidly moving parts, therefore its scene flow can be represented as a combination of the rigid motion of these individual parts. Building upon this observation, we propose to generate pseudo scene flow labels for self-supervised learning through piecewise rigid motion estimation, in which the source point cloud is decomposed into local regions and each region is treated as rigid. By rigidly aligning each region with its potential counterpart in the target point cloud, we obtain a region-specific rigid transformation to generate its pseudo flow labels. To mitigate the impact of potential outliers on label generation, when solving the rigid registration for each region, we alternately perform three steps: establishing point correspondences, measuring the confidence for the correspondences, and updating the rigid transformation based on the correspondences and their confidence. As a result, confident correspondences will dominate label generation and a validity mask will be derived for the generated pseudo labels. By using the pseudo labels together with their validity mask for supervision, models can be trained in a self-supervised manner. Extensive experiments on FlyingThings3D and KITTI datasets demonstrate that our method achieves new state-of-the-art performance in self-supervised scene flow learning, without any ground truth scene flow for supervision, even performing better than some supervised counterparts. Additionally, our method is further extended to class-agnostic motion prediction and significantly outperforms previous state-of-the-art self-supervised methods on nuScenes dataset.
Neural Radiance Fields (NeRF) has received much attention recently due to its impressive capability to represent 3D scene and synthesize novel view images. Existing works usually assume that the input images are captured by a global shutter camera. Thus, rolling shutter (RS) images cannot be trivially applied to an off-the-shelf NeRF algorithm for novel view synthesis. Rolling shutter effect would also affect the accuracy of the camera pose estimation (e.g. via COLMAP), which further prevents the success of NeRF algorithm with RS images. In this paper, we propose Unrolling Shutter Bundle Adjusted Neural Radiance Fields (USB-NeRF). USB-NeRF is able to correct rolling shutter distortions and recover accurate camera motion trajectory simultaneously under the framework of NeRF, by modeling the physical image formation process of a RS camera. Experimental results demonstrate that USB-NeRF achieves better performance compared to prior works, in terms of RS effect removal, novel view image synthesis as well as camera motion estimation. Furthermore, our algorithm can also be used to recover high-fidelity high frame-rate global shutter video from a sequence of RS images.
Learned B-frame video compression aims to adopt bi-directional motion estimation and motion compensation (MEMC) coding for middle frame reconstruction. However, previous learned approaches often directly extend neural P-frame codecs to B-frame relying on bi-directional optical-flow estimation or video frame interpolation. They suffer from inaccurate quantized motions and inefficient motion compensation. To address these issues, we propose a simple yet effective structure called Interpolation-driven B-frame Video Compression (IBVC). Our approach only involves two major operations: video frame interpolation and artifact reduction compression. IBVC introduces a bit-rate free MEMC based on interpolation, which avoids optical-flow quantization and additional compression distortions. Later, to reduce duplicate bit-rate consumption and focus on unaligned artifacts, a residual guided masking encoder is deployed to adaptively select the meaningful contexts with interpolated multi-scale dependencies. In addition, a conditional spatio-temporal decoder is proposed to eliminate location errors and artifacts instead of using MEMC coding in other methods. The experimental results on B-frame coding demonstrate that IBVC has significant improvements compared to the relevant state-of-the-art methods. Meanwhile, our approach can save bit rates compared with the random access (RA) configuration of H.266 (VTM). The code will be available at https://github.com/ruhig6/IBVC.
3D motion estimation from cine cardiac magnetic resonance (CMR) images is important for the assessment of cardiac function and the diagnosis of cardiovascular diseases. Current state-of-the art methods focus on estimating dense pixel-/voxel-wise motion fields in image space, which ignores the fact that motion estimation is only relevant and useful within the anatomical objects of interest, e.g., the heart. In this work, we model the heart as a 3D mesh consisting of epi- and endocardial surfaces. We propose a novel learning framework, DeepMesh, which propagates a template heart mesh to a subject space and estimates the 3D motion of the heart mesh from CMR images for individual subjects. In DeepMesh, the heart mesh of the end-diastolic frame of an individual subject is first reconstructed from the template mesh. Mesh-based 3D motion fields with respect to the end-diastolic frame are then estimated from 2D short- and long-axis CMR images. By developing a differentiable mesh-to-image rasterizer, DeepMesh is able to leverage 2D shape information from multiple anatomical views for 3D mesh reconstruction and mesh motion estimation. The proposed method estimates vertex-wise displacement and thus maintains vertex correspondences between time frames, which is important for the quantitative assessment of cardiac function across different subjects and populations. We evaluate DeepMesh on CMR images acquired from the UK Biobank. We focus on 3D motion estimation of the left ventricle in this work. Experimental results show that the proposed method quantitatively and qualitatively outperforms other image-based and mesh-based cardiac motion tracking methods.
Scene-aware global human motion forecasting is critical for manifold applications, including virtual reality, robotics, and sports. The task combines human trajectory and pose forecasting within the provided scene context, which represents a significant challenge. So far, only Mao et al. NeurIPS'22 have addressed scene-aware global motion, cascading the prediction of future scene contact points and the global motion estimation. They perform the latter as the end-to-end forecasting of future trajectories and poses. However, end-to-end contrasts with the coarse-to-fine nature of the task and it results in lower performance, as we demonstrate here empirically. We propose a STAGed contact-aware global human motion forecasting STAG, a novel three-stage pipeline for predicting global human motion in a 3D environment. We first consider the scene and the respective human interaction as contact points. Secondly, we model the human trajectory forecasting within the scene, predicting the coarse motion of the human body as a whole. The third and last stage matches a plausible fine human joint motion to complement the trajectory considering the estimated contacts. Compared to the state-of-the-art (SoA), STAG achieves a 1.8% and 16.2% overall improvement in pose and trajectory prediction, respectively, on the scene-aware GTA-IM dataset. A comprehensive ablation study confirms the advantages of staged modeling over end-to-end approaches. Furthermore, we establish the significance of a newly proposed temporal counter called the "time-to-go", which tells how long it is before reaching scene contact and endpoints. Notably, STAG showcases its ability to generalize to datasets lacking a scene and achieves a new state-of-the-art performance on CMU-Mocap, without leveraging any social cues. Our code is released at: https://github.com/L-Scofano/STAG
Mobile autonomy relies on the precise perception of dynamic environments. Robustly tracking moving objects in 3D world thus plays a pivotal role for applications like trajectory prediction, obstacle avoidance, and path planning. While most current methods utilize LiDARs or cameras for Multiple Object Tracking (MOT), the capabilities of 4D imaging radars remain largely unexplored. Recognizing the challenges posed by radar noise and point sparsity in 4D radar data, we introduce RaTrack, an innovative solution tailored for radar-based tracking. Bypassing the typical reliance on specific object types and 3D bounding boxes, our method focuses on motion segmentation and clustering, enriched by a motion estimation module. Evaluated on the View-of-Delft dataset, RaTrack showcases superior tracking precision of moving objects, largely surpassing the performance of the state of the art. We release our code and model at https://github.com/LJacksonPan/RaTrack.
Objective. A phased or a curvilinear array produces ultrasound (US) images with a sector field of view (FOV), which inherently exhibits spatially-varying image resolution with inferior quality in the far zone and towards the two sides azimuthally. Sector US images with improved spatial resolutions are favorable for accurate quantitative analysis of large and dynamic organs, such as the heart. Therefore, this study aims to translate US images with spatially-varying resolution to ones with less spatially-varying resolution. CycleGAN has been a prominent choice for unpaired medical image translation; however, it neither guarantees structural consistency nor preserves backscattering patterns between input and generated images for unpaired US images. Approach. To circumvent this limitation, we propose a constrained CycleGAN (CCycleGAN), which directly performs US image generation with unpaired images acquired by different ultrasound array probes. In addition to conventional adversarial and cycle-consistency losses of CycleGAN, CCycleGAN introduces an identical loss and a correlation coefficient loss based on intrinsic US backscattered signal properties to constrain structural consistency and backscattering patterns, respectively. Instead of post-processed B-mode images, CCycleGAN uses envelope data directly obtained from beamformed radio-frequency signals without any other non-linear postprocessing. Main Results. In vitro phantom results demonstrate that CCycleGAN successfully generates images with improved spatial resolution as well as higher peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) compared with benchmarks. Significance. CCycleGAN-generated US images of the in vivo human beating heart further facilitate higher quality heart wall motion estimation than benchmarks-generated ones, particularly in deep regions.
Temporal echocardiography image registration is a basis for clinical quantifications such as cardiac motion estimation, myocardial strain assessments, and stroke volume quantifications. In past studies, deep learning image registration (DLIR) has shown promising results and is consistently accurate and precise, requiring less computational time. We propose that a greater focus on the warped moving image's anatomic plausibility and image quality can support robust DLIR performance. Further, past implementations have focused on adult echocardiography, and there is an absence of DLIR implementations for fetal echocardiography. We propose a framework that combines three strategies for DLIR in both fetal and adult echo: (1) an anatomic shape-encoded loss to preserve physiological myocardial and left ventricular anatomical topologies in warped images; (2) a data-driven loss that is trained adversarially to preserve good image texture features in warped images; and (3) a multi-scale training scheme of a data-driven and anatomically constrained algorithm to improve accuracy. Our tests show that good anatomical topology and image textures are strongly linked to shape-encoded and data-driven adversarial losses. They improve different aspects of registration performance in a non-overlapping way, justifying their combination. Despite fundamental distinctions between adult and fetal echo images, we show that these strategies can provide excellent registration results in both adult and fetal echocardiography using the publicly available CAMUS adult echo dataset and our private multi-demographic fetal echo dataset. Our approach outperforms traditional non-DL gold standard registration approaches, including Optical Flow and Elastix. Registration improvements could be translated to more accurate and precise clinical quantification of cardiac ejection fraction, demonstrating a potential for translation.