Search

We address the challenging problem of learning motion representations using deep models for video recognition. To this end, we make use of attention modules that learn to highlight regions in the video and aggregate features for recognition. Specifically, we propose to leverage output attention maps as a vehicle to transfer the learned representation from a motion (flow) network to an RGB network. We systematically study the design of attention modules, and develop a novel method for attention distillation. Our method is evaluated on major action benchmarks, and consistently improves the performance of the baseline RGB network by a significant margin. Moreover, we demonstrate that our attention maps can leverage motion cues in learning to identify the location of actions in video frames. We believe our method provides a step towards learning motion-aware representations in deep models. Our project page is available at https://aptx4869lm.github.io/AttentionDistillation/

Eye contact is a crucial element of non-verbal communication that signifies interest, attention, and participation in social interactions. As a result, measures of eye contact arise in a variety of applications such as the assessment of the social communication skills of children at risk for developmental disorders such as autism, or the analysis of turn-taking and social roles during group meetings. However, the automated measurement of visual attention during naturalistic social interactions is challenging due to the difficulty of estimating a subject's looking direction from video. This paper proposes a novel approach to eye contact detection during adult-child social interactions in which the adult wears a point-of-view camera which captures an egocentric view of the child's behavior. By analyzing the child's face regions and inferring their head pose we can accurately identify the onset and duration of the child's looks to their social partner's eyes. We introduce the Pose-Implicit CNN, a novel deep learning architecture that predicts eye contact while implicitly estimating the head pose. We present a fully automated system for eye contact detection that solves the sub-problems of end-to-end feature learning and pose estimation using deep neural networks. To train our models, we use a dataset comprising 22 hours of 156 play session videos from over 100 children, half of whom are diagnosed with Autism Spectrum Disorder. We report an overall precision of 0.76, recall of 0.80, and an area under the precision-recall curve of 0.79, all of which are significant improvements over existing methods., Comment: Published in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT) Volume 1. Winner of IMWUT Distinguished Paper Award

We describe a novel cross-modal embedding space for actions, named Action2Vec, which combines linguistic cues from class labels with spatio-temporal features derived from video clips. Our approach uses a hierarchical recurrent network to capture the temporal structure of video features. We train our embedding using a joint loss that combines classification accuracy with similarity to Word2Vec semantics. We evaluate Action2Vec by performing zero shot action recognition and obtain state of the art results on three standard datasets. In addition, we present two novel analogy tests which quantify the extent to which our joint embedding captures distributional semantics. This is the first joint embedding space to combine verbs and action videos, and the first to be thoroughly evaluated with respect to its distributional semantics.

In this paper, we provide a modern synthesis of the classic inverse compositional algorithm for dense image alignment. We first discuss the assumptions made by this well-established technique, and subsequently propose to relax these assumptions by incorporating data-driven priors into this model. More specifically, we unroll a robust version of the inverse compositional algorithm and replace multiple components of this algorithm using more expressive models whose parameters we train in an end-to-end fashion from data. Our experiments on several challenging 3D rigid motion estimation tasks demonstrate the advantages of combining optimization with learning-based techniques, outperforming the classic inverse compositional algorithm as well as data-driven image-to-pose regression approaches., Comment: Paper accepted at CVPR 2019, oral presentation. Code is available at https://github.com/lvzhaoyang/DeeperInverseCompositionalAlgorithm

In this paper we present a framework for combining deep learning-based road detection, particle filters, and Model Predictive Control (MPC) to drive aggressively using only a monocular camera, IMU, and wheel speed sensors. This framework uses deep convolutional neural networks combined with LSTMs to learn a local cost map representation of the track in front of the vehicle. A particle filter uses this dynamic observation model to localize in a schematic map, and MPC is used to drive aggressively using this particle filter based state estimate. We show extensive real world testing results, and demonstrate reliable operation of the vehicle at the friction limits on a complex dirt track. We reach speeds above 27 mph (12 m/s) on a dirt track with a 105 foot (32m) long straight using our 1:5 scale test vehicle. A video of these results can be found at https://www.youtube.com/watch?v=5ALIK-z-vUg

Automatic generation of textual video descriptions that are time-aligned with video content is a long-standing goal in computer vision. The task is challenging due to the difficulty of bridging the semantic gap between the visual and natural language domains. This paper addresses the task of automatically generating an alignment between a set of instructions and a first person video demonstrating an activity. The sparse descriptions and ambiguity of written instructions create significant alignment challenges. The key to our approach is the use of egocentric cues to generate a concise set of action proposals, which are then matched to recipe steps using object recognition and computational linguistic techniques. We obtain promising results on both the Extended GTEA Gaze+ dataset and the Bristol Egocentric Object Interactions Dataset.

This article presents AutoRally, a 1$:$5 scale robotics testbed for autonomous vehicle research. AutoRally is designed for robustness, ease of use, and reproducibility, so that a team of two people with limited knowledge of mechanical engineering, electrical engineering, and computer science can construct and then operate the testbed to collect real world autonomous driving data in whatever domain they wish to study. Complete documentation to construct and operate the platform is available online along with tutorials, example controllers, and a driving dataset collected at the Georgia Tech Autonomous Racing Facility. Offline estimation algorithms are used to determine parameters for physics-based dynamics models using an adaptive limited memory joint state unscented Kalman filter. Online vehicle state estimation using a factor graph optimization scheme and a convolutional neural network for semantic segmentation of drivable surface are presented. All algorithms are tested with real world data from the fleet of six AutoRally robots at the Georgia Tech Autonomous Racing Facility tracks, and serve as a demonstration of the robot$'$s capabilities.

Inner product-based convolution has been a central component of convolutional neural networks (CNNs) and the key to learning visual representations. Inspired by the observation that CNN-learned features are naturally decoupled with the norm of features corresponding to the intra-class variation and the angle corresponding to the semantic difference, we propose a generic decoupled learning framework which models the intra-class variation and semantic difference independently. Specifically, we first reparametrize the inner product to a decoupled form and then generalize it to the decoupled convolution operator which serves as the building block of our decoupled networks. We present several effective instances of the decoupled convolution operator. Each decoupled operator is well motivated and has an intuitive geometric interpretation. Based on these decoupled operators, we further propose to directly learn the operator from data. Extensive experiments show that such decoupled reparameterization renders significant performance gain with easier convergence and stronger robustness., Comment: CVPR 2018 (Spotlight)

Estimation of 3D motion in a dynamic scene from a temporal pair of images is a core task in many scene understanding problems. In real world applications, a dynamic scene is commonly captured by a moving camera (i.e., panning, tilting or hand-held), increasing the task complexity because the scene is observed from different view points. The main challenge is the disambiguation of the camera motion from scene motion, which becomes more difficult as the amount of rigidity observed decreases, even with successful estimation of 2D image correspondences. Compared to other state-of-the-art 3D scene flow estimation methods, in this paper we propose to \emph{learn} the rigidity of a scene in a supervised manner from a large collection of dynamic scene data, and directly infer a rigidity mask from two sequential images with depths. With the learned network, we show how we can effectively estimate camera motion and projected scene flow using computed 2D optical flow and the inferred rigidity mask. For training and testing the rigidity network, we also provide a new semi-synthetic dynamic scene dataset (synthetic foreground objects with a real background) and an evaluation split that accounts for the percentage of observed non-rigid pixels. Through our evaluation we show the proposed framework outperforms current state-of-the-art scene flow estimation methods in challenging dynamic scenes., Comment: This work is accepted at European Conference on Computer Vision 2018. Project page (with the video): http://research.nvidia.com/publication/2018-09_Learning-Rigidity-in The codes will be released at https://github.com/NVlabs/learningrigidity

In this paper, we make an important step towards the black-box machine teaching by considering the cross-space machine teaching, where the teacher and the learner use different feature representations and the teacher can not fully observe the learner's model. In such scenario, we study how the teacher is still able to teach the learner to achieve faster convergence rate than the traditional passive learning. We propose an active teacher model that can actively query the learner (i.e., make the learner take exams) for estimating the learner's status and provably guide the learner to achieve faster convergence. The sample complexities for both teaching and query are provided. In the experiments, we compare the proposed active teacher with the omniscient teacher and verify the effectiveness of the active teacher model., Comment: Published in ICML 2018

Estimating the head pose of a person is a crucial problem that has a large amount of applications such as aiding in gaze estimation, modeling attention, fitting 3D models to video and performing face alignment. Traditionally head pose is computed by estimating some keypoints from the target face and solving the 2D to 3D correspondence problem with a mean human head model. We argue that this is a fragile method because it relies entirely on landmark detection performance, the extraneous head model and an ad-hoc fitting step. We present an elegant and robust way to determine pose by training a multi-loss convolutional neural network on 300W-LP, a large synthetically expanded dataset, to predict intrinsic Euler angles (yaw, pitch and roll) directly from image intensities through joint binned pose classification and regression. We present empirical tests on common in-the-wild pose benchmark datasets which show state-of-the-art results. Additionally we test our method on a dataset usually used for pose estimation using depth and start to close the gap with state-of-the-art depth pose methods. We open-source our training and testing code as well as release our pre-trained models., Comment: Accepted to Computer Vision and Pattern Recognition Workshops (CVPRW), 2018 IEEE Conference on. IEEE, 2018

Face detection is a very important task and a necessary pre-processing step for many applications such as facial landmark detection, pose estimation, sentiment analysis and face recognition. Not only is face detection an important pre-processing step in computer vision applications but also in computational psychology, behavioral imaging and other fields where researchers might not be initiated in computer vision frameworks and state-of-the-art detection applications. A large part of existing research that includes face detection as a pre-processing step uses existing out-of-the-box detectors such as the HoG-based dlib and the OpenCV Haar face detector which are no longer state-of-the-art - they are primarily used because of their ease of use and accessibility. We introduce Dockerface, a very accurate Faster R-CNN face detector in a Docker container which requires no training and is easy to install and use.

Gaze tracking is an important technology as the system can give information about a person from what and where the person is seeing. There have been many attempts to make robust and accurate gaze trackers using either monitor or wearable devices. However, those contraptions often require fine individual calibration per session and/or require a person wearing a device, which may not be suitable for certain situations. In this paper, we propose a robust and a completely noninvasive gaze tracking system that involves neither complex calibrations nor the use of wearable devices. We achieve this via direct eye reflection analysis by building a real-time system that effectively enables it. We also show several interesting applications for our system including experiments with young children.

We present a framework for vision-based model predictive control (MPC) for the task of aggressive, high-speed autonomous driving. Our approach uses deep convolutional neural networks to predict cost functions from input video which are directly suitable for online trajectory optimization with MPC. We demonstrate the method in a high speed autonomous driving scenario, where we use a single monocular camera and a deep convolutional neural network to predict a cost map of the track in front of the vehicle. Results are demonstrated on a 1:5 scale autonomous vehicle given the task of high speed, aggressive driving., Comment: 11 pages, 7 figures

Safe autonomous vehicles must be able to predict and react to the drivers around them. Previous control methods rely heavily on pre-computation and are unable to react to dynamic events as they unfold in real-time. In this paper, we extend Model Predictive Path Integral Control (MPPI) using differential game theory and introduce Best-Response MPPI (BR-MPPI) for real-time multi-vehicle interactions. Experimental results are presented using two AutoRally platforms in a racing format with BR-MPPI competing against a skilled human driver at the Georgia Tech Autonomous Racing Facility., Comment: 8 pages, 7 figures

We present an information theoretic approach to stochastic optimal control problems that can be used to derive general sampling based optimization schemes. This new mathematical method is used to develop a sampling based model predictive control algorithm. We apply this information theoretic model predictive control (IT-MPC) scheme to the task of aggressive autonomous driving around a dirt test track, and compare its performance to a model predictive control version of the cross-entropy method., Comment: 20 pages, 12 figures, submitted to Transactions on Robotics (T-RO)

In this paper, we consider the problem of machine teaching, the inverse problem of machine learning. Different from traditional machine teaching which views the learners as batch algorithms, we study a new paradigm where the learner uses an iterative algorithm and a teacher can feed examples sequentially and intelligently based on the current performance of the learner. We show that the teaching complexity in the iterative case is very different from that in the batch case. Instead of constructing a minimal training set for learners, our iterative machine teaching focuses on achieving fast convergence in the learner model. Depending on the level of information the teacher has from the learner model, we design teaching algorithms which can provably reduce the number of teaching examples and achieve faster convergence than learning without teachers. We also validate our theoretical findings with extensive experiments on different data distribution and real image datasets., Comment: Published in ICML 2017

Approximate Bayesian Computation (ABC) is a framework for performing likelihood-free posterior inference for simulation models. Stochastic Variational inference (SVI) is an appealing alternative to the inefficient sampling approaches commonly used in ABC. However, SVI is highly sensitive to the variance of the gradient estimators, and this problem is exacerbated by approximating the likelihood. We draw upon recent advances in variance reduction for SV and likelihood-free inference using deterministic simulations to produce low variance gradient estimators of the variational lower-bound. By then exploiting automatic differentiation libraries we can avoid nearly all model-specific derivations. We demonstrate performance on three problems and compare to existing SVI algorithms. Our results demonstrate the correctness and efficiency of our algorithm.

Data-driven approaches for edge detection have proven effective and achieve top results on modern benchmarks. However, all current data-driven edge detectors require manual supervision for training in the form of hand-labeled region segments or object boundaries. Specifically, human annotators mark semantically meaningful edges which are subsequently used for training. Is this form of strong, high-level supervision actually necessary to learn to accurately detect edges? In this work we present a simple yet effective approach for training edge detectors without human supervision. To this end we utilize motion, and more specifically, the only input to our method is noisy semi-dense matches between frames. We begin with only a rudimentary knowledge of edges (in the form of image gradients), and alternate between improving motion estimation and edge detection in turn. Using a large corpus of video data, we show that edge detectors trained using our unsupervised scheme approach the performance of the same methods trained with full supervision (within 3-5%). Finally, we show that when using a deep network for the edge detector, our approach provides a novel pre-training scheme for object detection., Comment: Camera ready version for CVPR 2016

Whole-arm tactile sensing enables a robot to sense contact and infer contact properties across its entire arm. Within this paper, we demonstrate that using data-driven methods, a humanoid robot can infer mechanical properties of objects from contact with its forearm during a simple reaching motion. A key issue is the extent to which the performance of data-driven methods can generalize to robot actions that differ from those used during training. To investigate this, we developed an idealized physics-based lumped element model of a robot with a compliant joint making contact with an object. Using this physics-based model, we performed experiments with varied robot, object and environment parameters. We also collected data from a tactile-sensing forearm on a real robot as it made contact with various objects during a simple reaching motion with varied arm velocities and joint stiffnesses. The robot used one nearest neighbor classifiers (1-NN), hidden Markov models (HMMs), and long short-term memory (LSTM) networks to infer two object properties (hard vs. soft and moved vs. unmoved) based on features of time-varying tactile sensor data (maximum force, contact area, and contact motion). We found that, in contrast to 1-NN, the performance of LSTMs (with sufficient data availability) and multivariate HMMs successfully generalized to new robot motions with distinct velocities and joint stiffnesses. Compared to single features, using multiple features gave the best results for both experiments with physics-based models and a real-robot., Comment: This updated version [v2] was accepted to the International Journal of Humanoid Robotics, Special Issue on Tactile Sensing for Manipulation : New Progress and Challenges

In this paper we provide an extensive evaluation of fixation prediction and salient object segmentation algorithms as well as statistics of major datasets. Our analysis identifies serious design flaws of existing salient object benchmarks, called the dataset design bias, by over emphasizing the stereotypical concepts of saliency. The dataset design bias does not only create the discomforting disconnection between fixations and salient object segmentation, but also misleads the algorithm designing. Based on our analysis, we propose a new high quality dataset that offers both fixation and salient object segmentation ground-truth. With fixations and salient object being presented simultaneously, we are able to bridge the gap between fixations and salient objects, and propose a novel method for salient object segmentation. Finally, we report significant benchmark progress on three existing datasets of segmenting salient objects, Comment: 15 pages, 8 figures. Conference version was accepted by CVPR 2014