Scene understanding under low-light conditions is a challenging problem. This is due to the small number of photons captured by the camera and the resulting low signal-to-noise ratio (SNR). Single-photon cameras (SPCs) are an emerging sensing modality that are capable of capturing images with high sensitivity. Despite having minimal read-noise, images captured by SPCs in photon-starved conditions still suffer from strong shot noise, preventing reliable scene inference. We propose photon scale-space a collection of high-SNR images spanning a wide range of photons-per-pixel (PPP) levels (but same scene content) as guides to train inference model on low photon flux images. We develop training techniques that push images with different illumination levels closer to each other in feature representation space. The key idea is that having a spectrum of different brightness levels during training enables effective guidance, and increases robustness to shot noise even in extreme noise cases. Based on the proposed approach, we demonstrate, via simulations and real experiments with a SPAD camera, high-performance on various inference tasks such as image classification and monocular depth estimation under ultra low-light, down to < 1 PPP.
We propose a novel dense mapping framework for sparse visual SLAM systems which leverages a compact scene representation. State-of-the-art sparse visual SLAM systems provide accurate and reliable estimates of the camera trajectory and locations of landmarks. While these sparse maps are useful for localization, they cannot be used for other tasks such as obstacle avoidance or scene understanding. In this paper we propose a dense mapping framework to complement sparse visual SLAM systems which takes as input the camera poses, keyframes and sparse points produced by the SLAM system and predicts a dense depth image for every keyframe. We build on CodeSLAM and use a variational autoencoder (VAE) which is conditioned on intensity, sparse depth and reprojection error images from sparse SLAM to predict an uncertainty-aware dense depth map. The use of a VAE then enables us to refine the dense depth images through multi-view optimization which improves the consistency of overlapping frames. Our mapper runs in a separate thread in parallel to the SLAM system in a loosely coupled manner. This flexible design allows for integration with arbitrary metric sparse SLAM systems without delaying the main SLAM process. Our dense mapper can be used not only for local mapping but also globally consistent dense 3D reconstruction through TSDF fusion. We demonstrate our system running with ORB-SLAM3 and show accurate dense depth estimation which could enable applications such as robotics and augmented reality.
It is difficult to collect data on a large scale in a monocular depth estimation because the task requires the simultaneous acquisition of RGB images and depths. Data augmentation is thus important to this task. However, there has been little research on data augmentation for tasks such as monocular depth estimation, where the transformation is performed pixel by pixel. In this paper, we propose a data augmentation method, called CutDepth. In CutDepth, part of the depth is pasted onto an input image during training. The method extends variations data without destroying edge features. Experiments objectively and subjectively show that the proposed method outperforms conventional methods of data augmentation. The estimation accuracy is improved with CutDepth even though there are few training data at long distances.
We integrate sparse radar data into a monocular depth estimation model and introduce a novel preprocessing method for reducing the sparseness and limited field of view provided by radar. We explore the intrinsic error of different radar modalities and show our proposed method results in more data points with reduced error. We further propose a novel method for estimating dense depth maps from monocular 2D images and sparse radar measurements using deep learning based on the deep ordinal regression network by Fu et al. Radar data are integrated by first converting the sparse 2D points to a height-extended 3D measurement and then including it into the network using a late fusion approach. Experiments are conducted on the nuScenes dataset. Our experiments demonstrate state-of-the-art performance in both day and night scenes.
In recent years, monocular depth estimation is applied to understand the surrounding 3D environment and has made great progress. However, there is an ill-posed problem on how to gain depth information directly from a single image. With the rapid development of deep learning, this problem is possible to be solved. Although more and more approaches are proposed one after another, most of existing methods inevitably lost details due to continuous downsampling when mapping from RGB space to depth space. To the end, we design a Multi-scale Features Network (MSFNet), which consists of Enhanced Diverse Attention (EDA) module and Upsample-Stage Fusion (USF) module. The EDA module employs the spatial attention method to learn significant spatial information, while USF module complements low-level detail information with high-level semantic information from the perspective of multi-scale feature fusion to improve the predicted effect. In addition, since the simple samples are always trained to a better effect first, the hard samples are difficult to converge. Therefore, we design a batch-loss to assign large loss factors to the harder samples in a batch. Experiments on NYU-Depth V2 dataset and KITTI dataset demonstrate that our proposed approach is more competitive with the state-of-the-art methods in both qualitative and quantitative evaluation.
With the rapid development of data-driven techniques, data has played an essential role in various computer vision tasks. Many realistic and synthetic datasets have been proposed to address different problems. However, there are lots of unresolved challenges: (1) the creation of dataset is usually a tedious process with manual annotations, (2) most datasets are only designed for a single specific task, (3) the modification or randomization of the 3D scene is difficult, and (4) the release of commercial 3D data may encounter copyright issue. This paper presents MINERVAS, a Massive INterior EnviRonments VirtuAl Synthesis system, to facilitate the 3D scene modification and the 2D image synthesis for various vision tasks. In particular, we design a programmable pipeline with Domain-Specific Language, allowing users to (1) select scenes from the commercial indoor scene database, (2) synthesize scenes for different tasks with customized rules, and (3) render various imagery data, such as visual color, geometric structures, semantic label. Our system eases the difficulty of customizing massive numbers of scenes for different tasks and relieves users from manipulating fine-grained scene configurations by providing user-controllable randomness using multi-level samplers. Most importantly, it empowers users to access commercial scene databases with millions of indoor scenes and protects the copyright of core data assets, e.g., 3D CAD models. We demonstrate the validity and flexibility of our system by using our synthesized data to improve the performance on different kinds of computer vision tasks.
Monocular depth estimation (MDE) is a fundamental task in many applications such as scene understanding and reconstruction. However, most of the existing methods rely on accurately labeled datasets. A weakly-supervised framework based on attention nested U-net (ANU) named as ANUW is introduced in this paper for cases with wrong labels. The ANUW is trained end-to-end to convert an input single RGB image into a depth image. It consists of a dense residual network structure, an adaptive weight channel attention (AWCA) module, a patch second non-local (PSNL) module and a soft label generation method. The dense residual network is the main body of the network to encode and decode the input. The AWCA module can adaptively adjust the channel weights to extract important features. The PSNL module implements the spatial attention mechanism through a second-order non-local method. The proposed soft label generation method uses the prior knowledge of the dataset to produce soft labels to replace false ones. The proposed ANUW is trained on a defective monocular depth dataset and the trained model is tested on three public datasets, and the results demonstrate the superiority of ANUW in comparison with the state-of-the-art MDE methods.
Dense depth estimation and 3D reconstruction of a surgical scene are crucial steps in computer assisted surgery. Recent work has shown that depth estimation from a stereo images pair could be solved with convolutional neural networks. However, most recent depth estimation models were trained on datasets with per-pixel ground truth. Such data is especially rare for laparoscopic imaging, making it hard to apply supervised depth estimation to real surgical applications. To overcome this limitation, we propose SADepth, a new self-supervised depth estimation method based on Generative Adversarial Networks. It consists of an encoder-decoder generator and a discriminator to incorporate geometry constraints during training. Multi-scale outputs from the generator help to solve the local minima caused by the photometric reprojection loss, while the adversarial learning improves the framework generation quality. Extensive experiments on two public datasets show that SADepth outperforms recent state-of-the-art unsupervised methods by a large margin, and reduces the gap between supervised and unsupervised depth estimation in laparoscopic images.
We present an algorithm to estimate fast and accurate depth maps from light fields via a sparse set of depth edges and gradients. Our proposed approach is based around the idea that true depth edges are more sensitive than texture edges to local constraints, and so they can be reliably disambiguated through a bidirectional diffusion process. First, we use epipolar-plane images to estimate sub-pixel disparity at a sparse set of pixels. To find sparse points efficiently, we propose an entropy-based refinement approach to a line estimate from a limited set of oriented filter banks. Next, to estimate the diffusion direction away from sparse points, we optimize constraints at these points via our bidirectional diffusion method. This resolves the ambiguity of which surface the edge belongs to and reliably separates depth from texture edges, allowing us to diffuse the sparse set in a depth-edge and occlusion-aware manner to obtain accurate dense depth maps.
We introduce TransformerFusion, a transformer-based 3D scene reconstruction approach. From an input monocular RGB video, the video frames are processed by a transformer network that fuses the observations into a volumetric feature grid representing the scene; this feature grid is then decoded into an implicit 3D scene representation. Key to our approach is the transformer architecture that enables the network to learn to attend to the most relevant image frames for each 3D location in the scene, supervised only by the scene reconstruction task. Features are fused in a coarse-to-fine fashion, storing fine-level features only where needed, requiring lower memory storage and enabling fusion at interactive rates. The feature grid is then decoded to a higher-resolution scene reconstruction, using an MLP-based surface occupancy prediction from interpolated coarse-to-fine 3D features. Our approach results in an accurate surface reconstruction, outperforming state-of-the-art multi-view stereo depth estimation methods, fully-convolutional 3D reconstruction approaches, and approaches using LSTM- or GRU-based recurrent networks for video sequence fusion.
As cameras are increasingly deployed in new application domains such as autonomous driving, performing 3D object detection on monocular images becomes an important task for visual scene understanding. Recent advances on monocular 3D object detection mainly rely on the ``pseudo-LiDAR'' generation, which performs monocular depth estimation and lifts the 2D pixels to pseudo 3D points. However, depth estimation from monocular images, due to its poor accuracy, leads to inevitable position shift of pseudo-LiDAR points within the object. Therefore, the predicted bounding boxes may suffer from inaccurate location and deformed shape. In this paper, we present a novel neighbor-voting method that incorporates neighbor predictions to ameliorate object detection from severely deformed pseudo-LiDAR point clouds. Specifically, each feature point around the object forms their own predictions, and then the ``consensus'' is achieved through voting. In this way, we can effectively combine the neighbors' predictions with local prediction and achieve more accurate 3D detection. To further enlarge the difference between the foreground region of interest (ROI) pseudo-LiDAR points and the background points, we also encode the ROI prediction scores of 2D foreground pixels into the corresponding pseudo-LiDAR points. We conduct extensive experiments on the KITTI benchmark to validate the merits of our proposed method. Our results on the bird's eye view detection outperform the state-of-the-art performance by a large margin, especially for the ``hard'' level detection.
A deep learning-based monocular depth estimation (MDE) technique is proposed for selection of most informative frames (key frames) of an endoscopic video. In most of the cases, ground truth depth maps of polyps are not readily available and that is why the transfer learning approach is adopted in our method. An endoscopic modalities generally capture thousands of frames. In this scenario, it is quite important to discard low-quality and clinically irrelevant frames of an endoscopic video while the most informative frames should be retained for clinical diagnosis. In this view, a key-frame selection strategy is proposed by utilizing the depth information of polyps. In our method, image moment, edge magnitude, and key-points are considered for adaptively selecting the key frames. One important application of our proposed method could be the 3D reconstruction of polyps with the help of extracted key frames. Also, polyps are localized with the help of extracted depth maps.
Reconstructing the scene of robotic surgery from the stereo endoscopic video is an important and promising topic in surgical data science, which potentially supports many applications such as surgical visual perception, robotic surgery education and intra-operative context awareness. However, current methods are mostly restricted to reconstructing static anatomy assuming no tissue deformation, tool occlusion and de-occlusion, and camera movement. However, these assumptions are not always satisfied in minimal invasive robotic surgeries. In this work, we present an efficient reconstruction pipeline for highly dynamic surgical scenes that runs at 28 fps. Specifically, we design a transformer-based stereoscopic depth perception for efficient depth estimation and a light-weight tool segmentor to handle tool occlusion. After that, a dynamic reconstruction algorithm which can estimate the tissue deformation and camera movement, and aggregate the information over time is proposed for surgical scene reconstruction. We evaluate the proposed pipeline on two datasets, the public Hamlyn Centre Endoscopic Video Dataset and our in-house DaVinci robotic surgery dataset. The results demonstrate that our method can recover the scene obstructed by the surgical tool and handle the movement of camera in realistic surgical scenarios effectively at real-time speed.
Depth perception is fundamental for robots to understand the surrounding environment. As the view of cognitive neuroscience, visual depth perception methods are divided into three categories, namely binocular, active, and pictorial. The first two categories have been studied for decades in detail. However, research for the exploration of the third category is still in its infancy and has got momentum by the advent of deep learning methods in recent years. In cognitive neuroscience, it is known that pictorial depth perception mechanisms are dependent on the perception of seen objects. Inspired by this fact, in this thesis, we investigated the relation of perception of objects and depth estimation convolutional neural networks. For this purpose, we developed new network structures based on a simple depth estimation network that only used a single image at its input. Our proposed structures use both an image and a semantic label of the image as their input. We used semantic labels as the output of object perception. The obtained results of performance comparison between the developed network and original network showed that our novel structures can improve the performance of depth estimation by 52\% of relative error of distance in the examined cases. Most of the experimental studies were carried out on synthetic datasets that were generated by game engines to isolate the performance comparison from the effect of inaccurate depth and semantic labels of non-synthetic datasets. It is shown that particular synthetic datasets may be used for training of depth networks in cases that an appropriate dataset is not available. Furthermore, we showed that in these cases, usage of semantic labels improves the robustness of the network against domain shift from synthetic training data to non-synthetic test data.
This paper proposes two important contributions for conditional Generative Adversarial Networks (cGANs) to improve the wide variety of applications that exploit this architecture. The first main contribution is an analysis of cGANs to show that they are not explicitly conditional. In particular, it will be shown that the discriminator and subsequently the cGAN does not automatically learn the conditionality between inputs. The second contribution is a new method, called a contrario cGAN, that explicitly models conditionality for both parts of the adversarial architecture via a novel a contrario loss that involves training the discriminator to learn unconditional (adverse) examples. This leads to a novel type of data augmentation approach for GANs (a contrario learning) which allows to restrict the search space of the generator to conditional outputs using adverse examples. Extensive experimentation is carried out to evaluate the conditionality of the discriminator by proposing a probability distribution analysis. Comparisons with the cGAN architecture for different applications show significant improvements in performance on well known datasets including, semantic image synthesis, image segmentation, monocular depth prediction and "single label"-to-image using different metrics including Fr\'echet Inception Distance (FID), mean Intersection over Union (mIoU), Root Mean Square Error log (RMSE log) and Number of statistically-Different Bins (NDB).
Instance segmentation of images is an important tool for automated scene understanding. Neural networks are usually trained to optimize their overall performance in terms of accuracy. Meanwhile, in applications such as automated driving, an overlooked pedestrian seems more harmful than a falsely detected one. In this work, we present a false negative detection method for image sequences based on inconsistencies in time series of tracked instances given the availability of image sequences in online applications. As the number of instances can be greatly increased by this algorithm, we apply a false positive pruning using uncertainty estimates aggregated over instances. To this end, instance-wise metrics are constructed which characterize uncertainty and geometry of a given instance or are predicated on depth estimation. The proposed method serves as a post-processing step applicable to any neural network that can also be trained on single frames only. In our tests, we obtain an improved trade-off between false negative and false positive instances by our fused detection approach in comparison to the use of an ordinary score value provided by the instance segmentation network during inference.
We present OffRoadTranSeg, the first end-to-end framework for semi-supervised segmentation in unstructured outdoor environment using transformers and automatic data selection for labelling. The offroad segmentation is a scene understanding approach that is widely used in autonomous driving. The popular offroad segmentation method is to use fully connected convolution layers and large labelled data, however, due to class imbalance, there will be several mismatches and also some classes may not be detected. Our approach is to do the task of offroad segmentation in a semi-supervised manner. The aim is to provide a model where self supervised vision transformer is used to fine-tune offroad datasets with self-supervised data collection for labelling using depth estimation. The proposed method is validated on RELLIS-3D and RUGD offroad datasets. The experiments show that OffRoadTranSeg outperformed other state of the art models, and also solves the RELLIS-3D class imbalance problem.
We propose a new network architecture, the Fractal Pyramid Networks (PFNs) for pixel-wise prediction tasks as an alternative to the widely used encoder-decoder structure. In the encoder-decoder structure, the input is processed by an encoding-decoding pipeline that tries to get a semantic large-channel feature. Different from that, our proposed PFNs hold multiple information processing pathways and encode the information to multiple separate small-channel features. On the task of self-supervised monocular depth estimation, even without ImageNet pretrained, our models can compete or outperform the state-of-the-art methods on the KITTI dataset with much fewer parameters. Moreover, the visual quality of the prediction is significantly improved. The experiment of semantic segmentation provides evidence that the PFNs can be applied to other pixel-wise prediction tasks, and demonstrates that our models can catch more global structure information.
Camouflaged object detection (COD) aims to segment camouflaged objects hiding in the environment, which is challenging due to the similar appearance of camouflaged objects and their surroundings. Research in biology suggests depth can provide useful object localization cues for camouflaged object discovery. In this paper, we study the depth contribution for camouflaged object detection, where the depth maps are generated with existing monocular depth estimation (MDE) methods. Due to the domain gap between the MDE dataset and our COD dataset, the generated depth maps are not accurate enough to be directly used. We then introduce two solutions to avoid the noisy depth maps from dominating the training process. Firstly, we present an auxiliary depth estimation branch ("ADE"), aiming to regress the depth maps. We find that "ADE" is especially necessary for our "generated depth" scenario. Secondly, we introduce a multi-modal confidence-aware loss function via a generative adversarial network to weigh the contribution of depth for camouflaged object detection. Our extensive experiments on various camouflaged object detection datasets explain that the existing "sensor depth" based RGB-D segmentation techniques work poorly with "generated depth", and our proposed two solutions work cooperatively, achieving effective depth contribution exploration for camouflaged object detection.
Novel view synthesis is required in many robotic applications, such as VR teleoperation and scene reconstruction. Existing methods are often too slow for these contexts, cannot handle dynamic scenes, and are limited by their explicit depth estimation stage, where incorrect depth predictions can lead to large projection errors. Our proposed method runs in real time on live streaming data and avoids explicit depth estimation by efficiently warping input images into the target frame for a range of assumed depth planes. The resulting plane sweep volume (PSV) is directly fed into our network, which first estimates soft PSV masks in a self-supervised manner, and then directly produces the novel output view. This improves efficiency and performance on transparent, reflective, thin, and feature-less scene parts. FaDIV-Syn can perform both interpolation and extrapolation tasks at 540p in real-time and outperforms state-of-the-art extrapolation methods on the large-scale RealEstate10k dataset. We thoroughly evaluate ablations, such as removing the Soft-Masking network, training from fewer examples as well as generalization to higher resolutions and stronger depth discretization. Our implementation is available.