Unsupervised methods have showed promising results on monocular depth estimation. However, the training data must be captured in scenes without moving objects. To push the envelope of accuracy, recent methods tend to increase their model parameters. In this paper, an unsupervised learning framework is proposed to jointly predict monocular depth and complete 3D motion including the motions of moving objects and camera. (1) Recurrent modulation units are used to adaptively and iteratively fuse encoder and decoder features. This improves the single-image depth inference without overspending model parameters. (2) Instead of using a single set of filters for upsampling, multiple sets of filters are devised for the residual upsampling. This facilitates the learning of edge-preserving filters and leads to the improved performance. (3) A warping-based network is used to estimate a motion field of moving objects without using semantic priors. This breaks down the requirement of scene rigidity and allows to use general videos for the unsupervised learning. The motion field is further regularized by an outlier-aware training loss. Despite the depth model just uses a single image in test time and 2.97M parameters, it achieves state-of-the-art results on the KITTI and Cityscapes benchmarks.

Over four decades, the majority addresses the problem of optical flow estimation using variational methods. With the advance of machine learning, some recent works have attempted to address the problem using convolutional neural network (CNN) and have showed promising results. FlowNet2 [1] , the state-of-the-art CNN, requires over 160M parameters to achieve accurate flow estimation. Our LiteFlowNet2 outperforms FlowNet2 on Sintel and KITTI benchmarks, while being 25.3 times smaller in the model size and 3.1 times faster in the running speed. LiteFlowNet2 is built on the foundation laid by conventional methods and resembles the corresponding roles as data fidelity and regularization in variational methods. We compute optical flow in a spatial-pyramid formulation as SPyNet [2] but through a novel lightweight cascaded flow inference. It provides high flow estimation accuracy through early correction with seamless incorporation of descriptor matching. Flow regularization is used to ameliorate the issue of outliers and vague flow boundaries through feature-driven local convolutions. Our network also owns an effective structure for pyramidal feature extraction and embraces feature warping rather than image warping as practiced in FlowNet2 and SPyNet. Comparing to LiteFlowNet [3] , LiteFlowNet2 improves the optical flow accuracy on Sintel Clean by 23.3 percent, Sintel Final by 12.8 percent, KITTI 2012 by 19.6 percent, and KITTI 2015 by 18.8 percent, while being 2.2 times faster. Our network protocol and trained models are made publicly available on https://github.com/twhui/LiteFlowNet2.

We study the problem of distilling knowledge from a large deep teacher network to a much smaller student network for the task of road marking segmentation. In this work, we explore a novel knowledge distillation (KD) approach that can transfer `knowledge' on scene structure more effectively from a teacher to a student model. Our method is known as Inter-Region Affinity KD (IntRA-KD). It decomposes a given road scene image into different regions and represents each region as a node in a graph. An inter-region affinity graph is then formed by establishing pairwise relationships between nodes based on their similarity in feature distribution. To learn structural knowledge from the teacher network, the student is required to match the graph generated by the teacher. The proposed method shows promising results on three large-scale road marking segmentation benchmarks, i.e., ApolloScape, CULane and LLAMAS, by taking various lightweight models as students and ResNet-101 as the teacher. IntRA-KD consistently brings higher performance gains on all lightweight models, compared to previous distillation methods. Our code is available at https://github.com/ cardwing/Codes-for-IntRA-KD.

FlowNet2 [14], the state-of-the-art convolutional neural network (CNN) for optical flow estimation, requires over 160M parameters to achieve accurate flow estimation. In this paper we present an alternative network that attains performance on par with FlowNet2 on the challenging Sintel final pass and KITTI benchmarks, while being 30 times smaller in the model size and 1.36 times faster in the running speed. This is made possible by drilling down to architectural details that might have been missed in the current frameworks: (1) We present a more effective flow inference approach at each pyramid level through a lightweight cascaded network. It not only improves flow estimation accuracy through early correction, but also permits seamless incorporation of descriptor matching in our network. (2) We present a novel flow regularization layer to ameliorate the issue of outliers and vague flow boundaries by using a feature-driven local convolution. (3) Our network owns an effective structure for pyramidal feature extraction and embraces feature warping rather than image warping as practiced in FlowNet2. Our code and trained models are available at github.com/twhui/LiteFlowNet.

Low-cost RGB-D imaging system such as Kinect is widely utilized for dense 3D reconstruction. However, RGB-D system generally suffers from two main problems. The spatial resolution of the depth image is low. The depth image often contains numerous holes where no depth measurements are available. This can be due to bad infra-red reflectance properties of some objects in the scene. Since the spatial resolution of the color image is generally higher than that of the depth image, this paper introduces a new method to enhance the depth images captured by a moving RGB-D system using the depth cues from the induced optical flow. We not only fill the holes in the raw depth images, but also recover fine details of the imaged scene. We address the problem of depth image enhancement by minimizing an energy functional. In order to reduce the computational complexity, we have treated the textured and homogeneous regions in the color images differently. Experimental results on several RGB-D sequences are provided to show the effectiveness of the proposed method.

Depth maps from low-cost RGB-D system are generally noisy and not accurate enough. Holes often exist in the depth maps. Bilateral filter is commonly utilized to perform depth enhancement. However, it requires high computational time. Its texture transferring property also makes those boundaries between textured and homogeneous regions in the filtered depth map far from satisfactory. In this paper, we present a method to filter raw depth maps using a RGB-D guided filtering in a two-stage framework. Our method not only has a faster computational time than bilateral filter but also avoids the problem of over-texture transfer. We also use RGB-D frames to fill holes in the depth maps. This can effectively prevents depth bleeding artifacts.

In this paper, we address the problem of dense depth map generation from two successive image frames in a video. We first recover the camera motion from the observable normal flow pattern using our previously proposed apparent flow constraints. Once the camera motion is estimated, sparse depth data can be directly recovered from the flow pattern. We utilize a hierarchical approach to generate an initial dense depth map from the sparse depth data. This depth map is further enhanced through the refinement of the associated optical flow field in a variational framework. Experimental results show that the proposed method can provide high-quality depth maps. We also have a faster computational time than the conventional optical flow approach.

We address the problem of recovering camera motion from video data, which does not require the establishment of feature correspondences or computation of optical flows but from normal flows directly. We have designed an imaging system that has a wide field of view by fixating a number of cameras together to form an approximate spherical eye. With a substantially widened visual field, we discover that estimating the directions of translation and rotation components of the motion separately are possible and particularly efficient. In addition, the inherent ambiguities between translation and rotation also disappear. Magnitude of rotation is recovered subsequently. Experimental results on synthetic and real image data are provided. The results show that not only the accuracy of motion estimation is comparable to those of the state-of-the-art methods that require explicit feature correspondences or optical flows, but also a faster computation time.

We address the problem of motion recovery and depth map reconstruction from a sequence of stereo images. The proposed method utilizes normal flows which are directly observable in image sequence. No explicit correspondence establishment is required. We first determine the direction of translation and the full rotation for each of the cameras in the stereo system using an iterative algorithm. Then we recover the translation magnitude using a geometrical method. Depth map in metric scale is reconstructed once the motion parameters are estimated. Experimental results on the synthetic and the real image data show that the proposed method is effective in both motion determination and depth map reconstruction.

We present a direct method for estimating general camera motion from a rectified stereo image sequence using merely normal flows. We require neither point-to-point feature correspondences nor special assumption about the imaged scene. It has a closed-form expression, requiring no iterative computation. Experimental results on both synthetic and real image data are provided to demonstrate the performance of the proposed method.