We present a deep model that can accurately produce dense depth maps given an RGB image with known depth at a very sparse set of pixels. The model works simultaneously for both indoor/outdoor scenes and produces state-of-the-art dense depth maps at nearly real-time speeds on both the NYUv2 and KITTI datasets. We surpass the state-of-the-art for monocular depth estimation even with depth values for only 1 out of every ~10000 image pixels, and we outperform other sparse-to-dense depth methods at all sparsity levels. With depth values for 1/256 of the image pixels, we achieve a mean absolute error of less than 1% of actual depth on indoor scenes, comparable to the performance of consumer-grade depth sensor hardware. Our experiments demonstrate that it would indeed be possible to efficiently transform sparse depth measurements obtained using e.g. lower-power depth sensors or SLAM systems into high-quality dense depth maps.
This paper proposes a deep neural network (DNN) for piece-wise planar depthmap reconstruction from a single RGB image. While DNNs have brought remarkable progress to single-image depth prediction, piece-wise planar depthmap reconstruction requires a structured geometry representation, and has been a difficult task to master even for DNNs. The proposed end-to-end DNN learns to directly infer a set of plane parameters and corresponding plane segmentation masks from a single RGB image. We have generated more than 50,000 piece-wise planar depthmaps for training and testing from ScanNet, a large-scale RGBD video database. Our qualitative and quantitative evaluations demonstrate that the proposed approach outperforms baseline methods in terms of both plane segmentation and depth estimation accuracy. To the best of our knowledge, this paper presents the first end-to-end neural architecture for piece-wise planar reconstruction from a single RGB image. Code and data are available at https://github.com/art-programmer/PlaneNet.
Deep neural networks are applied to a wide range of problems in recent years. In this work, Convolutional Neural Network (CNN) is applied to the problem of determining the depth from a single camera image (monocular depth). Eight different networks are designed to perform depth estimation, each of them suitable for a feature level. Networks with different pooling sizes determine different feature levels. After designing a set of networks, these models may be combined into a single network topology using graph optimization techniques. This "Semi Parallel Deep Neural Network (SPDNN)" eliminates duplicated common network layers, and can be further optimized by retraining to achieve an improved model compared to the individual topologies. In this study, four SPDNN models are trained and have been evaluated at 2 stages on the KITTI dataset. The ground truth images in the first part of the experiment are provided by the benchmark, and for the second part, the ground truth images are the depth map results from applying a state-of-the-art stereo matching method. The results of this evaluation demonstrate that using post-processing techniques to refine the target of the network increases the accuracy of depth estimation on individual mono images. The second evaluation shows that using segmentation data alongside the original data as the input can improve the depth estimation results to a point where performance is comparable with stereo depth estimation. The computational time is also discussed in this study.
We revisit the problem of visual depth estimation in the context of autonomous vehicles. Despite the progress on monocular depth estimation in recent years, we show that the gap between monocular and stereo depth accuracy remains large$-$a particularly relevant result due to the prevalent reliance upon monocular cameras by vehicles that are expected to be self-driving. We argue that the challenges of removing this gap are significant, owing to fundamental limitations of monocular vision. As a result, we focus our efforts on depth estimation by stereo. We propose a novel semi-supervised learning approach to training a deep stereo neural network, along with a novel architecture containing a machine-learned argmax layer and a custom runtime (that will be shared publicly) that enables a smaller version of our stereo DNN to run on an embedded GPU. Competitive results are shown on the KITTI 2015 stereo dataset. We also evaluate the recent progress of stereo algorithms by measuring the impact upon accuracy of various design criteria.
With the technological advancements of aerial imagery and accurate 3d reconstruction of urban environments, more and more attention has been paid to the automated analyses of urban areas. In our work, we examine two important aspects that allow live analysis of building structures in city models given oblique aerial imagery, namely automatic building extraction with convolutional neural networks (CNNs) and selective real-time depth estimation from aerial imagery. We use transfer learning to train the Faster R-CNN method for real-time deep object detection, by combining a large ground-based dataset for urban scene understanding with a smaller number of images from an aerial dataset. We achieve an average precision (AP) of about 80% for the task of building extraction on a selected evaluation dataset. Our evaluation focuses on both dataset-specific learning and transfer learning. Furthermore, we present an algorithm that allows for multi-view depth estimation from aerial imagery in real-time. We adopt the semi-global matching (SGM) optimization strategy to preserve sharp edges at object boundaries. In combination with the Faster R-CNN, it allows a selective reconstruction of buildings, identified with regions of interest (RoIs), from oblique aerial imagery.
The goal of our work is to complete the depth channel of an RGB-D image. Commodity-grade depth cameras often fail to sense depth for shiny, bright, transparent, and distant surfaces. To address this problem, we train a deep network that takes an RGB image as input and predicts dense surface normals and occlusion boundaries. Those predictions are then combined with raw depth observations provided by the RGB-D camera to solve for depths for all pixels, including those missing in the original observation. This method was chosen over others (e.g., inpainting depths directly) as the result of extensive experiments with a new depth completion benchmark dataset, where holes are filled in training data through the rendering of surface reconstructions created from multiview RGB-D scans. Experiments with different network inputs, depth representations, loss functions, optimization methods, inpainting methods, and deep depth estimation networks show that our proposed approach provides better depth completions than these alternatives.
While an increasing interest in deep models for single-image depth estimation methods can be observed, established schemes for their evaluation are still limited. We propose a set of novel quality criteria, allowing for a more detailed analysis by focusing on specific characteristics of depth maps. In particular, we address the preservation of edges and planar regions, depth consistency, and absolute distance accuracy. In order to employ these metrics to evaluate and compare state-of-the-art single-image depth estimation approaches, we provide a new high-quality RGB-D dataset. We used a DSLR camera together with a laser scanner to acquire high-resolution images and highly accurate depth maps. Experimental results show the validity of our proposed evaluation protocol.
It is an exciting task to recover the scene's 3d-structure and camera pose from the video sequence. Most of the current solutions divide it into two parts, monocular depth recovery and camera pose estimation. The monocular depth recovery is often studied as an independent part, and a better depth estimation is used to solve the pose. While camera pose is still estimated by traditional SLAM (Simultaneous Localization And Mapping) methods in most cases. The use of unsupervised method for monocular depth recovery and pose estimation has benefited from the study of [1] and achieved good results. In this paper, we improve the method of [1]. Our emphasis is laid on the improvement of the idea and related theory, introducing a more reasonable inter frame constraints and finally synthesize the camera trajectory with inter frame pose estimation in the unified world coordinate system. And our results get better performance.
The unsupervised depth estimation is the recent trend by utilizing the binocular stereo images to get rid of depth map ground truth. In unsupervised depth computation, the disparity images are generated by training the CNN with an image reconstruction loss. In this paper, a dual CNN based model is presented for unsupervised depth estimation with 6 losses (DNM6) with individual CNN for each view to generate the corresponding disparity map. The proposed dual CNN model is also extended with 12 losses (DNM12) by utilizing the cross disparities. The presented DNM6 and DNM12 models are experimented over KITTI driving and Cityscapes urban database and compared with the recent state-of-the-art result of unsupervised depth estimation. The code is available at: https://github.com/ishmav16/Dual-CNN-Models-for-Unsupervised-Monocular-Depth-Estimation.
Objects often occlude each other in scenes; Inferring their appearance beyond their visible parts plays an important role in scene understanding, depth estimation, object interaction and manipulation. In this paper, we study the challenging problem of completing the appearance of occluded objects. Doing so requires knowing which pixels to paint (segmenting the invisible parts of objects) and what color to paint them (generating the invisible parts). Our proposed novel solution, SeGAN, jointly optimizes for both segmentation and generation of the invisible parts of objects. Our experimental results show that: (a) SeGAN can learn to generate the appearance of the occluded parts of objects; (b) SeGAN outperforms state-of-the-art segmentation baselines for the invisible parts of objects; (c) trained on synthetic photo realistic images, SeGAN can reliably segment natural images; (d) by reasoning about occluder occludee relations, our method can infer depth layering.
Depth estimation and scene parsing are two particularly important tasks in visual scene understanding. In this paper we tackle the problem of simultaneous depth estimation and scene parsing in a joint CNN. The task can be typically treated as a deep multi-task learning problem [42]. Different from previous methods directly optimizing multiple tasks given the input training data, this paper proposes a novel multi-task guided prediction-and-distillation network (PAD-Net), which first predicts a set of intermediate auxiliary tasks ranging from low level to high level, and then the predictions from these intermediate auxiliary tasks are utilized as multi-modal input via our proposed multi-modal distillation modules for the final tasks. During the joint learning, the intermediate tasks not only act as supervision for learning more robust deep representations but also provide rich multi-modal information for improving the final tasks. Extensive experiments are conducted on two challenging datasets (i.e. NYUD-v2 and Cityscapes) for both the depth estimation and scene parsing tasks, demonstrating the effectiveness of the proposed approach.
We present ``just-in-time reconstruction" as real-time image-guided inpainting of a map with arbitrary scale and sparsity to generate a fully dense depth map for the image. In particular, our goal is to inpaint a sparse map --- obtained from either a monocular visual SLAM system or a sparse sensor --- using a single-view depth prediction network as a virtual depth sensor. We adopt a fairly standard approach to data fusion, to produce a fused depth map by performing inference over a novel fully-connected Conditional Random Field (CRF) which is parameterized by the input depth maps and their pixel-wise confidence weights. Crucially, we obtain the confidence weights that parameterize the CRF model in a data-dependent manner via Convolutional Neural Networks (CNNs) which are trained to model the conditional depth error distributions given each source of input depth map and the associated RGB image. Our CRF model penalises absolute depth error in its nodes and pairwise scale-invariant depth error in its edges, and the confidence-based fusion minimizes the impact of outlier input depth values on the fused result. We demonstrate the flexibility of our method by real-time inpainting of ORB-SLAM, Kinect, and LIDAR depth maps acquired both indoors and outdoors at arbitrary scale and varied amount of irregular sparsity.
Learning-based, single-view depth estimation often generalizes poorly to unseen datasets. While learning-based, two-frame depth estimation solves this problem to some extent by learning to match features across frames, it performs poorly at large depth where the uncertainty is high. There exists few learning-based, multi-view depth estimation methods. In this paper, we present a learning-based, multi-view dense depth map and ego-motion estimation method that uses Recurrent Neural Networks (RNN). Our model is designed for 3D reconstruction from video where the input frames are temporally correlated. It is generalizable to single- or two-view dense depth estimation. Compared to recent single- or two-view CNN-based depth estimation methods, our model leverages more views and achieves more accurate results, especially at large distances. Our method produces superior results to the state-of-the-art learning-based, single- or two-view depth estimation methods on both indoor and outdoor benchmark datasets. We also demonstrate that our method can even work on extremely difficult sequences, such as endoscopic video, where none of the assumptions (static scene, constant lighting, Lambertian reflection, etc.) from traditional 3D reconstruction methods hold.
Multi-task convolutional neural networks (CNNs) have shown impressive results for certain combinations of tasks, such as single-image depth estimation (SIDE) and semantic segmentation. This is achieved by pushing the network towards learning a robust representation that generalizes well to different atomic tasks. We extend this concept by adding auxiliary tasks, which are of minor relevance for the application, to the set of learned tasks. As a kind of additional regularization, they are expected to boost the performance of the ultimately desired main tasks. To study the proposed approach, we picked vision-based road scene understanding (RSU) as an exemplary application. Since multi-task learning requires specialized datasets, particularly when using extensive sets of tasks, we provide a multi-modal dataset for multi-task RSU, called synMT. More than 2.5 $\cdot$ 10^5 synthetic images, annotated with 21 different labels, were acquired from the video game Grand Theft Auto V (GTA V). Our proposed deep multi-task CNN architecture was trained on various combination of tasks using synMT. The experiments confirmed that auxiliary tasks can indeed boost network performance, both in terms of final results and training time.
Pixel-wise losses, e.g., cross-entropy or L2, have been widely used in structured prediction tasks as a spatial extension of generic image classification or regression. However, its i.i.d. assumption neglects the structural regularity present in natural images. Various attempts have been made to incorporate structural reasoning mostly through structure priors in a cooperative way where co-occurring patterns are encouraged. We, on the other hand, approach this problem from an opposing angle and propose a new framework, Adversarial Structure Matching (ASM), for training such structured prediction networks via an adversarial process, in which we train a structure analyzer that provides the supervisory signals, the ASM loss. The structure analyzer is trained to maximize the ASM loss, or to emphasize recurring multi-scale hard negative structural mistakes among co-occurring patterns. On the contrary, the structured prediction network is trained to reduce those mistakes and is thus enabled to distinguish fine-grained structures. As a result, training structured prediction networks using ASM reduces contextual confusion among objects and improves boundary localization. We demonstrate that our ASM outperforms pixel-wise IID loss or structural prior GAN loss on three different structured prediction tasks: semantic segmentation, monocular depth estimation, and surface normal prediction.
We address the unsupervised learning of several interconnected problems in low-level vision: single view depth prediction, camera motion estimation, optical flow, and segmentation of a video into the static scene and moving regions. Our key insight is that these four fundamental vision problems are coupled through geometric constraints. Consequently, learning to solve them together simplifies the problem because the solutions can reinforce each other. We go beyond previous work by exploiting geometry more explicitly and segmenting the scene into static and moving regions. To that end, we introduce Competitive Collaboration, a framework that facilitates the coordinated training of multiple specialized neural networks to solve complex problems. Competitive Collaboration works much like expectation-maximization, but with neural networks that act as both competitors to explain pixels that correspond to static or moving regions, and as collaborators through a moderator that assigns pixels to be either static or independently moving. Our novel method integrates all these problems in a common framework and simultaneously reasons about the segmentation of the scene into moving objects and the static background, the camera motion, depth of the static scene structure, and the optical flow of moving objects. Our model is trained without any supervision and achieves state-of-the-art performance among joint unsupervised methods on all sub-problems.
Deep learning has emerged as a powerful artificial intelligence tool to interpret medical images for a growing variety of applications. However, the paucity of medical imaging data with high-quality annotations that is necessary for training such methods ultimately limits their performance. Medical data is challenging to acquire due to privacy issues, shortage of experts available for annotation, limited representation of rare conditions and cost. This problem has previously been addressed by using synthetically generated data. However, networks trained on synthetic data often fail to generalize to real data. Cinematic rendering simulates the propagation and interaction of light passing through tissue models reconstructed from CT data, enabling the generation of photorealistic images. In this paper, we present one of the first applications of cinematic rendering in deep learning, in which we propose to fine-tune synthetic data-driven networks using cinematically rendered CT data for the task of monocular depth estimation in endoscopy. Our experiments demonstrate that: (a) Convolutional Neural Networks (CNNs) trained on synthetic data and fine-tuned on photorealistic cinematically rendered data adapt better to real medical images and demonstrate more robust performance when compared to networks with no fine-tuning, (b) these fine-tuned networks require less training data to converge to an optimal solution, and (c) fine-tuning with data from a variety of photorealistic rendering conditions of the same scene prevents the network from learning patient-specific information and aids in generalizability of the model. Our empirical evaluation demonstrates that networks fine-tuned with cinematically rendered data predict depth with 56.87% less error for rendered endoscopy images and 27.49% less error for real porcine colon endoscopy images.
We present an unsupervised approach for learning to estimate three dimensional (3D) facial structure from a single image while also predicting 3D viewpoint transformations that match a desired pose and facial geometry. We achieve this by inferring the depth of facial keypoints of an input image in an unsupervised manner, without using any form of ground-truth depth information. We show how it is possible to use these depths as intermediate computations within a new backpropable loss to predict the parameters of a 3D affine transformation matrix that maps inferred 3D keypoints of an input face to the corresponding 2D keypoints on a desired target facial geometry or pose. Our resulting approach, called DepthNets, can therefore be used to infer plausible 3D transformations from one face pose to another, allowing faces to be frontalized, transformed into 3D models or even warped to another pose and facial geometry. Lastly, we identify certain shortcomings with our formulation, and explore adversarial image translation techniques as a post-processing step to re-synthesize complete head shots for faces re-targeted to different poses or identities.
We consider the problem of scaling deep generative shape models to high-resolution. Drawing motivation from the canonical view representation of objects, we introduce a novel method for the fast up-sampling of 3D objects in voxel space through networks that perform super-resolution on the six orthographic depth projections. This allows us to generate high-resolution objects with more efficient scaling than methods which work directly in 3D. We decompose the problem of 2D depth super-resolution into silhouette and depth prediction to capture both structure and fine detail. This allows our method to generate sharp edges more easily than an individual network. We evaluate our work on multiple experiments concerning high-resolution 3D objects, and show our system is capable of accurately predicting novel objects at resolutions as large as 512$\mathbf{\times}$512$\mathbf{\times}$512 -- the highest resolution reported for this task. We achieve state-of-the-art performance on 3D object reconstruction from RGB images on the ShapeNet dataset, and further demonstrate the first effective 3D super-resolution method.
Most existing algorithms for depth estimation from single monocular images need large quantities of metric groundtruth depths for supervised learning. We show that relative depth can be an informative cue for metric depth estimation and can be easily obtained from vast stereo videos. Acquiring metric depths from stereo videos is sometimes impracticable due to the absence of camera parameters. In this paper, we propose to improve the performance of metric depth estimation with relative depths collected from stereo movie videos using existing stereo matching algorithm. We introduce a new "Relative Depth in Stereo" (RDIS) dataset densely labelled with relative depths. We first pretrain a ResNet model on our RDIS dataset. Then we finetune the model on RGB-D datasets with metric ground-truth depths. During our finetuning, we formulate depth estimation as a classification task. This re-formulation scheme enables us to obtain the confidence of a depth prediction in the form of probability distribution. With this confidence, we propose an information gain loss to make use of the predictions that are close to ground-truth. We evaluate our approach on both indoor and outdoor benchmark RGB-D datasets and achieve state-of-the-art performance.