skip to main content

Title: DF-Net: Unsupervised Joint Learning of Depth and Flow Using Cross-Task Consistency
We present an unsupervised learning framework for simultaneously training single-view depth prediction and optical flow estimation models using unlabeled video sequences. Existing unsupervised methods often exploit brightness constancy and spatial smoothness priors to train depth or flow models. In this paper, we propose to leverage geometric consistency as additional supervisory signals. Our core idea is that for rigid regions we can use the predicted scene depth and camera motion to synthesize 2D optical flow by backprojecting the induced 3D scene flow. The discrepancy between the rigid flow (from depth prediction and camera motion) and the estimated flow (from optical flow model) allows us to impose a cross-task consistency loss. While all the networks are jointly optimized during training, they can be applied independently at test time. Extensive experiments demonstrate that our depth and flow models compare favorably with state-of-the-art unsupervised methods.  more » « less
Award ID(s):
Author(s) / Creator(s):
; ;
Date Published:
Journal Name:
European Conference on Computer Vision
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Visual odometry (VO) and single image depth estimation are critical for robot vision, 3D reconstruction, and camera pose estimation that can be applied to autonomous driving, map building, augmented reality and many other applications. Various supervised learning models have been proposed to train the VO or single image depth estimation framework for each targeted scene to improve the performance recently. However, little effort has been made to learn these separate tasks together without requiring the collection of a significant number of labels. This paper proposes a novel unsupervised learning approach to simultaneously perceive VO and single image depth estimation. In our framework, either of these tasks can benefit from each other through simultaneously learning these two tasks. We correlate these two tasks by enforcing depth consistency between VO and single image depth estimation. Based on the single image depth estimation, we can resolve the most common and challenging scaling issue of monocular VO. Meanwhile, through training from a sequence of images, VO can enhance the single image depth estimation accuracy. The effectiveness of our proposed method is demonstrated through extensive experiments compared with current state-of-the-art methods on the benchmark datasets. 
    more » « less
  2. Robots are active agents that operate in dynamic scenarios with noisy sensors. Predictions based on these noisy sensor measurements often lead to errors and can be unreliable. To this end, roboticists have used fusion methods using multiple observations. Lately, neural networks have dominated the accuracy charts for perception-driven predictions for robotic decision-making and often lack uncertainty metrics associated with the predictions. Here, we present a mathematical formulation to obtain the heteroscedastic aleatoric uncertainty of any arbitrary distribution without prior knowledge about the data. The approach has no prior assumptions about the prediction labels and is agnostic to network architecture. Furthermore, our class of networks, Ajna, adds minimal computation and requires only a small change to the loss function while training neural networks to obtain uncertainty of predictions, enabling real-time operation even on resource-constrained robots. In addition, we study the informational cues present in the uncertainties of predicted values and their utility in the unification of common robotics problems. In particular, we present an approach to dodge dynamic obstacles, navigate through a cluttered scene, fly through unknown gaps, and segment an object pile, without computing depth but rather using the uncertainties of optical flow obtained from a monocular camera with onboard sensing and computation. We successfully evaluate and demonstrate the proposed Ajna network on four aforementioned common robotics and computer vision tasks and show comparable results to methods directly using depth. Our work demonstrates a generalized deep uncertainty method and demonstrates its utilization in robotics applications.

    more » « less
  3. Current deep neural network approaches for camera pose estimation rely on scene structure for 3D motion estimation, but this decreases the robustness and thereby makes cross-dataset generalization difficult. In contrast, classical approaches to structure from motion estimate 3D motion utilizing optical flow and then compute depth. Their accuracy, however, depends strongly on the quality of the optical flow. To avoid this issue, direct methods have been proposed, which separate 3D motion from depth estimation, but compute 3D motion using only image gradients in the form of normal flow. In this paper, we introduce a network NFlowNet, for normal flow estimation which is used to enforce robust and direct constraints. In particular, normal flow is used to estimate relative camera pose based on the cheirality (depth positivity) constraint. We achieve this by formulating the optimization problem as a differentiable cheirality layer, which allows for end-to-end learning of camera pose. We perform extensive qualitative and quantitative evaluation of the proposed DiffPoseNet’s sensitivity to noise and its generalization across datasets. We compare our approach to existing state-of-the-art methods on KITTI, TartanAir, and TUM-RGBD datasets. 
    more » « less
  4. Disentangling the sources of visual motion in a dynamic scene during self-movement or ego motion is important for autonomous navigation and tracking. In the dynamic image segments of a video frame containing independently moving objects, optic flow relative to the next frame is the sum of the motion fields generated due to camera and object motion. The traditional ego-motion estimation methods assume the scene to be static, and the recent deep learning-based methods do not separate pixel velocities into object- and ego-motion components. We propose a learning-based approach to predict both ego-motion parameters and object-motion field (OMF) from image sequences using a convolutional autoencoder while being robust to variations due to the unconstrained scene depth. This is achieved by: 1) training with continuous ego-motion constraints that allow solving for ego-motion parameters independently of depth and 2) learning a sparsely activated overcomplete ego-motion field (EMF) basis set, which eliminates the irrelevant components in both static and dynamic segments for the task of ego-motion estimation. In order to learn the EMF basis set, we propose a new differentiable sparsity penalty function that approximates the number of nonzero activations in the bottleneck layer of the autoencoder and enforces sparsity more effectively than L1- and L2-norm-based penalties. Unlike the existing direct ego-motion estimation methods, the predicted global EMF can be used to extract OMF directly by comparing it against the optic flow. Compared with the state-of-the-art baselines, the proposed model performs favorably on pixelwise object- and ego-motion estimation tasks when evaluated on real and synthetic data sets of dynamic scenes. 
    more » « less
  5. Synthetic data is highly useful for training machine learning systems performing image-based 3D reconstruction, as synthetic data has applications in both extending existing generalizable datasets and being tailored to train neural networks for specific learning tasks of interest. In this paper, we introduce and utilize a synthetic data generation suite capable of generating data given existing 3D scene models as input. Specifically, we use our tool to generate image sequences for use with Multi-View Stereo (MVS), moving a camera through the virtual space according to user-chosen camera parameters. We evaluate how the given camera parameters and type of 3D environment affect how applicable the generated image sequences are to the MVS task using five pre-trained neural networks on image sequences generated from three different 3D scene datasets. We obtain generated predictions for each combination of parameter value and input image sequence, using standard error metrics to analyze the differences in depth predictions on image sequences across 3D datasets, parameters, and networks. Among other results, we find that camera height and vertical camera viewing angle are the parameters that cause the most variation in depth prediction errors on these image sequences. 
    more » « less