We present a new learning-based method for multi-frame depth estimation from a color video, which is a fundamental problem in scene understanding, robot navigation or handheld 3D reconstruction. While recent learning-based methods estimate depth at high accuracy, 3D point clouds exported from their depth maps often fail to preserve important geometric feature (e.g., corners, edges, planes) of man-made scenes. Widely-used pixel-wise depth errors do not specifically penalize inconsistency on these features. These inaccuracies are particularly severe when subsequent depth reconstructions are accumulated in an attempt to scan a full environment with man-made objects with this kind of features. Our depth estimation algorithm therefore introduces a Combined Normal Map (CNM) constraint, which is designed to better preserve high-curvature features and global planar regions. In order to further improve the depth estimation accuracy, we introduce a new occlusion-aware strategy that aggregates initial depth predictions from multiple adjacent views into one final depth map and one occlusion probability map for the current reference view. Our method outperforms the state-of-the-art in terms of depth estimation accuracy, and preserves essential geometric features of man-made indoor scenes much better than other algorithms.
In this work we study the mutual benefits of two common computer vision tasks, self-supervised depth estimation and semantic segmentation from images. For example, to help unsupervised monocular depth estimation, constraints from semantic segmentation has been explored implicitly such as sharing and transforming features. In contrast, we propose to explicitly measure the border consistency between segmentation and depth and minimize it in a greedy manner by iteratively supervising the network towards a locally optimal solution. Partially this is motivated by our observation that semantic segmentation even trained with limited ground truth (200 images of KITTI) can offer more accurate border than that of any (monocular or stereo) image-based depth estimation. Through extensive experiments, our proposed approach advances the state of the art on unsupervised monocular depth estimation in the KITTI.
Whole understanding of the surroundings is paramount to autonomous systems. Recent works have shown that deep neural networks can learn geometry (depth) and motion (optical flow) from a monocular video without any explicit supervision from ground truth annotations, particularly hard to source for these two tasks. In this paper, we take an additional step toward holistic scene understanding with monocular cameras by learning depth and motion alongside with semantics, with supervision for the latter provided by a pre-trained network distilling proxy ground truth images. We address the three tasks jointly by a) a novel training protocol based on knowledge distillation and self-supervision and b) a compact network architecture which enables efficient scene understanding on both power hungry GPUs and low-power embedded platforms. We thoroughly assess the performance of our framework and show that it yields state-of-the-art results for monocular depth estimation, optical flow and motion segmentation.
Almost all previous deep learning-based multi-view stereo (MVS) approaches focus on improving reconstruction quality. Besides quality, efficiency is also a desirable feature for MVS in real scenarios. Towards this end, this paper presents a Fast-MVSNet, a novel sparse-to-dense coarse-to-fine framework, for fast and accurate depth estimation in MVS. Specifically, in our Fast-MVSNet, we first construct a sparse cost volume for learning a sparse and high-resolution depth map. Then we leverage a small-scale convolutional neural network to encode the depth dependencies for pixels within a local region to densify the sparse high-resolution depth map. At last, a simple but efficient Gauss-Newton layer is proposed to further optimize the depth map. On one hand, the high-resolution depth map, the data-adaptive propagation method and the Gauss-Newton layer jointly guarantee the effectiveness of our method. On the other hand, all modules in our Fast-MVSNet are lightweight and thus guarantee the efficiency of our approach. Besides, our approach is also memory-friendly because of the sparse depth representation. Extensive experimental results show that our method is 5$\times$ and 14$\times$ faster than Point-MVSNet and R-MVSNet, respectively, while achieving comparable or even better results on the challenging Tanks and Temples dataset as well as the DTU dataset. Code is available at https://github.com/svip-lab/FastMVSNet.
In the current monocular depth research, the dominant approach is to employ unsupervised training on large datasets, driven by warped photometric consistency. Such approaches lack robustness and are unable to generalize to challenging domains such as nighttime scenes or adverse weather conditions where assumptions about photometric consistency break down. We propose DeFeat-Net (Depth & Feature network), an approach to simultaneously learn a cross-domain dense feature representation, alongside a robust depth-estimation framework based on warped feature consistency. The resulting feature representation is learned in an unsupervised manner with no explicit ground-truth correspondences required. We show that within a single domain, our technique is comparable to both the current state of the art in monocular depth estimation and supervised feature representation learning. However, by simultaneously learning features, depth and motion, our technique is able to generalize to challenging domains, allowing DeFeat-Net to outperform the current state-of-the-art with around 10% reduction in all error measures on more challenging sequences such as nighttime driving.
With the mass-market adoption of dual-camera mobile phones, leveraging stereo information in computer vision has become increasingly important. Current state-of-the-art methods utilize learning-based algorithms, where the amount and quality of training samples heavily influence results. Existing stereo image datasets are limited either in size or subject variety. Hence, algorithms trained on such datasets do not generalize well to scenarios encountered in mobile photography. We present Holopix50k, a novel in-the-wild stereo image dataset, comprising 49,368 image pairs contributed by users of the Holopix mobile social platform. In this work, we describe our data collection process and statistically compare our dataset to other popular stereo datasets. We experimentally show that using our dataset significantly improves results for tasks such as stereo super-resolution and self-supervised monocular depth estimation. Finally, we showcase practical applications of our dataset to motivate novel works and use cases. The Holopix50k dataset is available at http://github.com/leiainc/holopix50k
This paper reports a new continuous 3D loss function for learning depth from monocular images. The dense depth prediction from a monocular image is supervised using sparse LIDAR points, which enables us to leverage available open source datasets with camera-LIDAR sensor suites during training. Currently, accurate and affordable range sensor is not readily available. Stereo cameras and LIDARs measure depth either inaccurately or sparsely/costly. In contrast to the current point-to-point loss evaluation approach, the proposed 3D loss treats point clouds as continuous objects; therefore, it compensates for the lack of dense ground truth depth due to LIDAR's sparsity measurements. We applied the proposed loss in three state-of-the-art monocular depth prediction approaches DORN, BTS, and Monodepth2. Experimental evaluation shows that the proposed loss improves the depth prediction accuracy and produces point-clouds with more consistent 3D geometric structures compared with all tested baselines, implying the benefit of the proposed loss on general depth prediction networks. A video demo of this work is available at https://youtu.be/5HL8BjSAY4Y.
Multi-view stereo (MVS) is the golden mean between the accuracy of active depth sensing and the practicality of monocular depth estimation. Cost volume based approaches employing 3D convolutional neural networks (CNNs) have considerably improved the accuracy of MVS systems. However, this accuracy comes at a high computational cost which impedes practical adoption. Distinct from cost volume approaches, we propose an efficient depth estimation approach by first (a) detecting and evaluating descriptors for interest points, then (b) learning to match and triangulate a small set of interest points, and finally (c) densifying this sparse set of 3D points using CNNs. An end-to-end network efficiently performs all three steps within a deep learning framework and trained with intermediate 2D image and 3D geometric supervision, along with depth supervision. Crucially, our first step complements pose estimation using interest point detection and descriptor learning. We demonstrate state-of-the-art results on depth estimation with lower compute for different scene lengths. Furthermore, our method generalizes to newer environments and the descriptors output by our network compare favorably to strong baselines. Code is available at https://github.com/magicleap/DELTAS
Disparity/depth estimation from sequences of stereo images is an important element in 3D vision. Owing to occlusions, imperfect settings and homogeneous luminance, accurate estimate of depth remains a challenging problem. Targetting view synthesis, we propose a novel learning-based framework making use of dilated convolution, densely connected convolutional modules, compact decoder and skip connections. The network is shallow but dense, so it is fast and accurate. Two additional contributions - a non-linear adjustment of the depth resolution and the introduction of a projection loss, lead to reduction of estimation error by up to 20% and 25% respectively. The results show that our network outperforms state-of-the-art methods with an average improvement in accuracy of depth estimation and view synthesis by approximately 45% and 34% respectively. Where our method generates comparable quality of estimated depth, it performs 10 times faster than those methods.
Differentiable image sampling in the form of backward warping has seen broad adoption in tasks like depth estimation and optical flow prediction. In contrast, how to perform forward warping has seen less attention, partly due to additional challenges such as resolving the conflict of mapping multiple pixels to the same target location in a differentiable way. We propose softmax splatting to address this paradigm shift and show its effectiveness on the application of frame interpolation. Specifically, given two input frames, we forward-warp the frames and their feature pyramid representations based on an optical flow estimate using softmax splatting. In doing so, the softmax splatting seamlessly handles cases where multiple source pixels map to the same target location. We then use a synthesis network to predict the interpolation result from the warped representations. Our softmax splatting allows us to not only interpolate frames at an arbitrary time but also to fine tune the feature pyramid and the optical flow. We show that our synthesis approach, empowered by softmax splatting, achieves new state-of-the-art results for video frame interpolation.
Light Field (LF) offers unique advantages such as post-capture refocusing and depth estimation, but low-light conditions limit these capabilities. To restore low-light LFs we should harness the geometric cues present in different LF views, which is not possible using single-frame low-light enhancement techniques. We, therefore, propose a deep neural network for Low-Light Light Field (L3F) restoration, which we refer to as L3Fnet. The proposed L3Fnet not only performs the necessary visual enhancement of each LF view but also preserves the epipolar geometry across views. We achieve this by adopting a two-stage architecture for L3Fnet. Stage-I looks at all the LF views to encode the LF geometry. This encoded information is then used in Stage-II to reconstruct each LF view. To facilitate learning-based techniques for low-light LF imaging, we collected a comprehensive LF dataset of various scenes. For each scene, we captured four LFs, one with near-optimal exposure and ISO settings and the others at different levels of low-light conditions varying from low to extreme low-light settings. The effectiveness of the proposed L3Fnet is supported by both visual and numerical comparisons on this dataset. To further analyze the performance of low-light reconstruction methods, we also propose an L3F-wild dataset that contains LF captured late at night with almost zero lux values. No ground truth is available in this dataset. To perform well on the L3F-wild dataset, any method must adapt to the light level of the captured scene. To do this we propose a novel pre-processing block that makes L3Fnet robust to various degrees of low-light conditions. Lastly, we show that L3Fnet can also be used for low-light enhancement of single-frame images, despite it being engineered for LF data. We do so by converting the single-frame DSLR image into a form suitable to L3Fnet, which we call as pseudo-LF.
In autonomous driving, monocular sequences contain lots of information. Monocular depth estimation, camera ego-motion estimation and optical flow estimation in consecutive frames are high-profile concerns recently. By analyzing tasks above, pixels in the middle frame are modeled into three parts: the rigid region, the non-rigid region, and the occluded region. In joint unsupervised training of depth and pose, we can segment the occluded region explicitly. The occlusion information is used in unsupervised learning of depth, pose and optical flow, as the image reconstructed by depth-pose and optical flow will be invalid in occluded regions. A less-than-mean mask is designed to further exclude the mismatched pixels interfered with by motion or illumination change in the training of depth and pose networks. This method is also used to exclude some trivial mismatched pixels in the training of the optical flow network. Maximum normalization is proposed for depth smoothness term to restrain depth degradation in textureless regions. In the occluded region, as depth and camera motion can provide more reliable motion estimation, they can be used to instruct unsupervised learning of optical flow. Our experiments in KITTI dataset demonstrate that the model based on three regions, full and explicit segmentation of the occlusion region, the rigid region, and the non-rigid region with corresponding unsupervised losses can improve performance on three tasks significantly. The source code is available at: https://github.com/guangmingw/DOPlearning.
We propose a learning-based network for depth map estimation from multi-view stereo (MVS) images. Our proposed network consists of three sub-networks: 1) a base network for initial depth map estimation from an unstructured stereo image pair, 2) a novel refinement network that leverages both photometric and geometric information, and 3) an attentional multi-view aggregation framework that enables efficient information exchange and integration among different stereo image pairs. The proposed network, called A-TVSNet, is evaluated on various MVS datasets and shows the ability to produce high quality depth map that outperforms competing approaches. Our code is available at https://github.com/daiszh/A-TVSNet.
Current methods for depth map prediction from monocular images tend to predict smooth, poorly localized contours for the occlusion boundaries in the input image. This is unfortunate as occlusion boundaries are important cues to recognize objects, and as we show, may lead to a way to discover new objects from scene reconstruction. To improve predicted depth maps, recent methods rely on various forms of filtering or predict an additive residual depth map to refine a first estimate. We instead learn to predict, given a depth map predicted by some reconstruction method, a 2D displacement field able to re-sample pixels around the occlusion boundaries into sharper reconstructions. Our method can be applied to the output of any depth estimation method, in an end-to-end trainable fashion. For evaluation, we manually annotated the occlusion boundaries in all the images in the test split of popular NYUv2-Depth dataset. We show that our approach improves the localization of occlusion boundaries for all state-of-the-art monocular depth estimation methods that we could evaluate, without degrading the depth accuracy for the rest of the images.
Self-supervised learning is showing great promise for monocular depth estimation, using geometry as the only source of supervision. Depth networks are indeed capable of learning representations that relate visual appearance to 3D properties by implicitly leveraging category-level patterns. In this work we investigate how to leverage more directly this semantic structure to guide geometric representation learning, while remaining in the self-supervised regime. Instead of using semantic labels and proxy losses in a multi-task approach, we propose a new architecture leveraging fixed pretrained semantic segmentation networks to guide self-supervised representation learning via pixel-adaptive convolutions. Furthermore, we propose a two-stage training process to overcome a common semantic bias on dynamic objects via resampling. Our method improves upon the state of the art for self-supervised monocular depth prediction over all pixels, fine-grained details, and per semantic categories.
The acquisition of light field images with high angular resolution is costly. Although many methods have been proposed to improve the angular resolution of a sparsely-sampled light field, they always focus on the light field with a small baseline, which is captured by a consumer light field camera. By making full use of the intrinsic \textit{geometry} information of light fields, in this paper we propose an end-to-end learning-based approach aiming at angularly super-resolving a sparsely-sampled light field with a large baseline. Our model consists of two learnable modules and a physically-based module. Specifically, it includes a depth estimation module for explicitly modeling the scene geometry, a physically-based warping for novel views synthesis, and a light field blending module specifically designed for light field reconstruction. Moreover, we introduce a novel loss function to promote the preservation of the light field parallax structure. Experimental results over various light field datasets including large baseline light field images demonstrate the significant superiority of our method when compared with state-of-the-art ones, i.e., our method improves the PSNR of the second best method up to 2 dB in average, while saves the execution time 48$\times$. In addition, our method preserves the light field parallax structure better.
We design a multiscopic vision system that utilizes a low-cost monocular RGB camera to acquire accurate depth estimation for robotic applications. Unlike multi-view stereo with images captured at unconstrained camera poses, the proposed system actively controls a robot arm with a mounted camera to capture a sequence of images in horizontally or vertically aligned positions with the same parallax. In this system, we combine the cost volumes for stereo matching between the reference image and the surrounding images to form a fused cost volume that is robust to outliers. Experiments on the Middlebury dataset and real robot experiments show that our obtained disparity maps are more accurate than two-frame stereo matching: the average absolute error is reduced by 50.2% in our experiments.
The completeness of 3D models is still a challenging problem in multi-view stereo (MVS) due to the unreliable photometric consistency in low-textured areas. Since low-textured areas usually exhibit strong planarity, planar models are advantageous to the depth estimation of low-textured areas. On the other hand, PatchMatch multi-view stereo is very efficient for its sampling and propagation scheme. By taking advantage of planar models and PatchMatch multi-view stereo, we propose a planar prior assisted PatchMatch multi-view stereo framework in this paper. In detail, we utilize a probabilistic graphical model to embed planar models into PatchMatch multi-view stereo and contribute a novel multi-view aggregated matching cost. This novel cost takes both photometric consistency and planar compatibility into consideration, making it suited for the depth estimation of both non-planar and planar regions. Experimental results demonstrate that our method can efficiently recover the depth information of extremely low-textured areas, thus obtaining high complete 3D models and achieving state-of-the-art performance.
Detecting and matching robust viewpoint-invariant keypoints is critical for visual SLAM and Structure-from-Motion. State-of-the-art learning-based methods generate training samples via homography adaptation to create 2D synthetic views with known keypoint matches from a single image. This approach, however, does not generalize to non-planar 3D scenes with illumination variations commonly seen in real-world videos. In this work, we propose self-supervised learning of depth-aware keypoints directly from unlabeled videos. We jointly learn keypoint and depth estimation networks by combining appearance and geometric matching via a differentiable structure-from-motion module based on Procrustean residual pose correction. We describe how our self-supervised keypoints can be integrated into state-of-the-art visual odometry frameworks for robust and accurate ego-motion estimation of autonomous vehicles in real-world conditions.
Camera calibration is an essential first step in setting up 3D Computer Vision systems. Commonly used parametric camera models are limited to a few degrees of freedom and thus often do not optimally fit to complex real lens distortion. In contrast, generic camera models allow for very accurate calibration due to their flexibility. Despite this, they have seen little use in practice. In this paper, we argue that this should change. We propose a calibration pipeline for generic models that is fully automated, easy to use, and can act as a drop-in replacement for parametric calibration, with a focus on accuracy. We compare our results to parametric calibrations. Considering stereo depth estimation and camera pose estimation as examples, we show that the calibration error acts as a bias on the results. We thus argue that in contrast to current common practice, generic models should be preferred over parametric ones whenever possible. To facilitate this, we released our calibration pipeline at https://github.com/puzzlepaint/camera_calibration, making both easy-to-use and accurate camera calibration available to everyone.