No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Cross-modal Learning by Hallucinating Missing Modalities in RGB-D Vision
ResearchGate has not been able to resolve any citations for this publication.
https://arxiv.org/abs/1803.10750
In this work, we propose a technique that tackles the video understanding problem under a realistic, demanding condition in which we have limited labeled data and partially observed training modalities. Common methods such as transfer learning do not take advantage of the rich information from extra modalities potentially available in the source domain dataset. On the other hand, previous work on cross-modality learning only focuses on a single domain or task. In this work, we propose a graph-based distillation method that incorporates rich privileged information from a large multi-modal dataset in the source domain, and shows an improved performance in the target domain where data is scarce. Leveraging both a large-scale dataset and its extra modalities, our method learns a better model for temporal action detection and action classification without needing to have access to these modalities during test time. We evaluate our approach on action classification and temporal action detection tasks, and show that our models achieve the state-of-the-art performance on the PKU-MMD and NTU RGB+D datasets.
In this work, we face the problem of unsupervised domain adaptation with a novel deep learning approach which leverages on our finding that entropy minimization is induced by the optimal alignment of second order statistics between source and target domains. We formally demonstrate this hypothesis and, aiming at achieving an optimal alignment in practical cases, we adopt a more principled strategy which, differently from the current Euclidean approaches, deploys alignment along geodesics. Our pipeline can be implemented by adding to the standard classification loss (on the labeled source domain), a source-to-target regularizer that is weighted in an unsupervised and data-driven fashion. We provide extensive experiments to assess the superiority of our framework on standard domain and modality adaptation benchmarks.
Recent works showed that Generative Adversarial Networks (GANs) can be successfully applied in unsupervised domain adaptation, where, given a labeled source dataset and an unlabeled target dataset, the goal is to train powerful classifiers for the target samples. In particular, it was shown that a GAN objective function can be used to learn target features indistinguishable from the source ones. In this work, we extend this framework by (i) forcing the learned feature extractor to be domain-invariant, and (ii) training it through data augmentation in the feature space, namely performing feature augmentation. While data augmentation in the image space is a well established technique in deep learning, feature augmentation has not yet received the same level of attention. We accomplish it by means of a feature generator trained by playing the GAN minimax game against source features. Results show that both enforcing domain-invariance and performing feature augmentation lead to superior or comparable performance to state-of-the-art results in several unsupervised domain adaptation benchmarks.
In video-based action recognition, viewpoint variations often pose major challenges because the same actions can appear different from different views. We use the complementary RGB and Depth information from the RGB-D cameras to address this problem. The proposed technique capitalizes on the spatio-temporal information available in the two data streams to the extract action features that are largely insensitive to the viewpoint variations. We use the RGB data to compute dense trajectories that are translated to viewpoint insensitive deep features under a non-linear knowledge transfer model. Similarly, the Depth stream is used to extract CNN-based view invariant features on which Fourier Temporal Pyramid is computed to incorporate the temporal information. The heterogeneous features from the two streams are combined and used as a dictionary to predict the label of the test samples. To that end, we propose a sparse-dense collaborative representation classification scheme that strikes a balance between the discriminative abilities of the dense and the sparse representations of the samples over the extracted heterogeneous dictionary.
Depth sensors open up possibilities of dealing with the human action recognition problem by providing 3D human skeleton data and depth images of the scene. Analysis of human actions based on 3D skeleton data has become popular recently, due to its robustness and view-invariant representation. However, the skeleton alone is insufficient to distinguish actions which involve human-object interactions. In this paper, we propose a deep model which efficiently models human-object interactions and intra-class variations under viewpoint changes. First, a human body-part model is introduced to transfer the depth appearances of body-parts to a shared view-invariant space. Second, an end-to-end learning framework is proposed which is able to effectively combine the view-invariant body-part representation from skeletal and depth images, and learn the relations between the human body-parts and the environmental objects, the interactions between different human body-parts, and the temporal structure of human actions. We have evaluated the performance of our proposed model against 15 existing techniques on two large benchmark human action recognition datasets including NTU RGB+D and UWA3DII. The Experimental results show that our technique provides a significant improvement over state-of-the-art methods.
Model distillation is an effective and widely used technique to transfer knowledge from a teacher to a student network. The typical application is to transfer from a powerful large network or ensemble to a small network, that is better suited to low-memory or fast execution requirements. In this paper, we present a deep mutual learning (DML) strategy where, rather than one way transfer between a static pre-defined teacher and a student, an ensemble of students learn collaboratively and teach each other throughout the training process. Our experiments show that a variety of network architectures benefit from mutual learning and achieve compelling results on CIFAR-100 recognition and Market-1501 person re-identification benchmarks. Surprisingly, it is revealed that no prior powerful teacher network is necessary -- mutual learning of a collection of simple student networks works, and moreover outperforms distillation from a more powerful yet static teacher.
In this paper we address the problem of semantic labeling of indoor scenes on RGB-D data. With the availability of RGB-D cameras, it is expected that additional depth measurement will improve the accuracy. Here we investigate a solution how to incorporate complementary depth information into a semantic segmentation framework by making use of convolutional neural networks (CNNs). Recently encoder-decoder type fully convolutional CNN architectures have achieved a great success in the field of semantic segmentation. Motivated by this observation we propose an encoder-decoder type network, where the encoder part is composed of two branches of networks that simultaneously extract features from RGB and depth images and fuse depth features into the RGB feature maps as the network goes deeper. Comprehensive experimental evaluations demonstrate that the proposed fusion-based architecture achieves competitive results with the state-of-the-art methods on the challenging SUN RGB-D benchmark obtaining 76.27% global accuracy , 48.30% average class accuracy and 37.29% average intersection-over-union score.
We present a variety of new architectural features and training procedures that we apply to the generative adversarial networks (GANs) framework. We focus on two applications of GANs: semi-supervised learning, and the generation of images that humans find visually realistic. Unlike most work on generative models, our primary goal is not to train a model that assigns high likelihood to test data, nor do we require the model to be able to learn well without using any labels. Using our new techniques, we achieve state-of-the-art results in semi-supervised classification on MNIST, CIFAR-10 and SVHN. The generated images are of high quality as confirmed by a visual Turing test: our model generates MNIST samples that humans cannot distinguish from real data, and CIFAR-10 samples that yield a human error rate of 21.3%. We also present ImageNet samples with unprecedented resolution and show that our methods enable the model to learn recognizable features of ImageNet classes.
This paper addresses the problem of estimating the depth map of a scene given a single RGB image. To model the ambiguous mapping between monocular images and depth maps, we leverage on deep learning capabilities and present a fully convolutional architecture encompassing residual learning. The proposed model is deeper than the current state of the art, but contains fewer parameters and requires less training data, while still outperforming all current CNN approaches aimed at the same task. We further present a novel way to efficiently learn feature map up-sampling within the network. For optimization we introduce the reverse Huber loss, particularly suited for the task at hand and driven by the value distributions commonly present in depth maps. The predictions are given by a single architecture, trained end-to-end, that does not rely on post-processing techniques, such as CRFs or other additional refinement steps.
Recent approaches in depth-based human activity analysis achieved outstanding performance and proved the effectiveness of 3D representation for classification of action classes. Currently available depth-based and RGB+D-based action recognition benchmarks have a number of limitations, including the lack of training samples, distinct class labels, camera views and variety of subjects. In this paper we introduce a large-scale dataset for RGB+D human action recognition with more than 56 thousand video samples and 4 million frames, collected from 40 distinct subjects. Our dataset contains 60 different action classes including daily, mutual, and health-related actions. In addition, we propose a new recurrent neural network structure to model the long-term temporal correlation of the features for each body part, and utilize them for better action classification. Experimental results show the advantages of applying deep learning methods over state-of-the-art hand-crafted features on the suggested cross-subject and cross-view evaluation criteria for our dataset. The introduction of this large scale dataset will enable the community to apply, develop and adapt various data-hungry learning techniques for the task of depth-based and RGB+D-based human activity analysis.
Most existing feature learning based methods for RGB-D object recognition either combine RGB and depth data in an undifferentiated manner from the outset, or learn features from color and depth separately, which do not adequately exploit different characteristics of the two modalities or utilize the shared relationship between the modalities. In this paper, we propose a general CNN based multi-modal learning framework for RGB-D object recognition. We first construct deep CNN layers for color and depth separately which are then connected with a carefully designed multimodal layer. This layer is designed to not only discover the most discriminative features for each modality, but is also able to harness the complementary relationship between the two modalities. The results of the multi-modal layer are backpropagated to update parameters of the CNN layers, and the multi-modal feature learning and the back-propagation are iteratively performed until convergence. Experimental results on two widely used RGB-D object datasets show that our method for general multi-modal learning achieves comparable performance to state-of-the-art methods specifically designed for RGB-D data.
We present a method for gesture detection and localisation based on
multi-scale and multi-modal deep learning. Each visual modality captures
spatial information at a particular spatial scale (such as motion of the upper
body or a hand), and the whole system operates at three temporal scales. Key to
our technique is a training strategy which exploits: i) careful initialization
of individual modalities; and ii) gradual fusion involving random dropping of
separate channels (dubbed ModDrop) for learning cross-modality correlations
while preserving uniqueness of each modality-specific representation. We
present experiments on the ChaLearn 2014 Looking at People Challenge gesture
recognition track, in which we placed first out of 17 teams. Fusing multiple
modalities at several spatial and temporal scales leads to a significant
increase in recognition rates, allowing the model to compensate for errors of
the individual classifiers as well as noise in the separate channels.
Futhermore, the proposed ModDrop training technique ensures robustness of the
classifier to missing signals in one or several channels to produce meaningful
predictions from any number of available modalities. In addition, we
demonstrate the applicability of the proposed fusion scheme to modalities of
arbitrary nature by experiments on the same dataset augmented with audio.
In this paper we study the problem of object detection for RGB-D images using
semantically rich image and depth features. We propose a new geocentric
embedding for depth images that encodes height above ground and angle with
gravity for each pixel in addition to the horizontal disparity. We demonstrate
that this geocentric embedding works better than using raw depth images for
learning feature representations with convolutional neural networks. Our final
object detection system achieves an average precision of 37.3%, which is a 56%
relative improvement over existing methods. We then focus on the task of
instance segmentation where we label pixels belonging to object instances found
by our detector. For this task, we propose a decision forest approach that
classifies pixels in the detection window as foreground or background using a
family of unary and binary tests that query shape and geocentric pose features.
Finally, we use the output from our object detectors in an existing superpixel
classification framework for semantic scene segmentation and achieve a 24%
relative improvement over current state-of-the-art for the object categories
that we study. We believe advances such as those represented in this paper will
facilitate the use of perception in fields like robotics.
In this paper, we characterize the noise in Kinect depth images based on multiple factors and introduce a uniform nomenclature for the types of noise. In the process, we briefly survey the noise models of Kinect and relate these to the factors of characterization. We also deal with the noise in multi-Kinect set-ups and summarize the techniques for the minimization of interference noise. Studies on noise in Kinect depth images are distributed over several publications and there is no comprehensive treatise on it. This paper, to the best of our knowledge, is the maiden attempt to characterize the noise behavior of Kinect depth images in a structured manner. The characterization would help to selectively eliminate noise from depth images either by filtering or by adopting appropriate methodologies for image capture. In addition to the characterization based on the results reported by others, we also conduct independent experiments in a number of cases to fill up the gaps in characterization and to validate the reported results.
Depth perception is the ability to see the three-dimensional volume of objects and the spatial layout of objects relative to one another and the viewer. Humans accomplish depth perception using a variety of cues, including some based on how the eyes and brain function in concert and others that rely on lawful regularities in the environment. The topics emphasized in this article are consistent with a current imbalance in scientific understanding, which is that much more is known about the specific cues involved than about how the brain uses these cues to achieve depth perception.
Both convolutional and recurrent operations are building blocks that process one local neighborhood at a time. In this paper, we present non-local operations as a generic family of building blocks for capturing long-range dependencies. Inspired by the classical non-local means method in computer vision, our non-local operation computes the response at a position as a weighted sum of the features at all positions. This building block can be plugged into many computer vision architectures. On the task of video classification, even without any bells and whistles, our non-local models can compete or outperform current competition winners on both Kinetics and Charades datasets. In static image recognition, our non-local models improve object detection/segmentation and pose estimation on the COCO suite of tasks. Code will be made available.
We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to 1/2 everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.
Rectified activation units (rectifiers) are essential for state-of-the-art neural networks. In this work, we study rectifier neural networks for image classification from two aspects. First, we propose a Parametric Rectified Linear Unit (PReLU) that generalizes the traditional rectified unit. PReLU improves model fitting with nearly zero extra computational cost and little overfitting risk. Second, we derive a robust initialization method that particularly considers the rectifier nonlinearities. This method enables us to train extremely deep rectified models directly from scratch and to investigate deeper or wider network architectures. Based on our PReLU networks (PReLU-nets), we achieve 4.94% top-5 test error on the ImageNet 2012 classification dataset. This is a 26% relative improvement over the ILSVRC 2014 winner (GoogLeNet, 6.66%). To our knowledge, our result is the first to surpass human-level performance (5.1%, Russakovsky et al.) on this visual recognition challenge.
In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning. We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning. Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks - demonstrating their applicability as general image representations.
A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.
We present an approach to interpret the major surfaces, objects, and support relations of an indoor scene from an RGBD image. Most existing work ignores physical interactions or is applied only to tidy rooms and hallways. Our goal is to parse typical, often messy, indoor scenes into floor, walls, supporting surfaces, and object regions, and to recover support relationships. One of our main interests is to better understand how 3D cues can best inform a structured 3D interpretation. We also contribute a novel integer programming formulation to infer physical support relations. We offer a new dataset of 1449 RGBD images, capturing 464 diverse indoor scenes, with detailed annotations. Our experiments demonstrate our ability to infer support relations in complex scenes and verify that our 3D scene cues and inferred support lead to better object segmentation.
Adversarial learning methods are a promising approach to training robust deep networks, and can generate complex samples across diverse domains. They also can improve recognition despite the presence of domain shift or dataset bias: several adversarial approaches to unsupervised domain adaptation have recently been introduced, which reduce the difference between the training and test domain distributions and thus improve generalization performance. Prior generative approaches show compelling visualizations, but are not optimal on discriminative tasks and can be limited to smaller shifts. Prior discriminative approaches could handle larger domain shifts, but imposed tied weights on the model and did not exploit a GAN-based loss. We first outline a novel generalized framework for adversarial adaptation, which subsumes recent state-of-the-art approaches as special cases, and we use this generalized view to better relate the prior approaches. We propose a previously unexplored instance of our general framework which combines discriminative modeling, untied weight sharing, and a GAN loss, which we call Adversarial Discriminative Domain Adaptation (ADDA). We show that ADDA is more effective yet considerably simpler than competing domain-adversarial methods, and demonstrate the promise of our approach by exceeding state-of-the-art unsupervised adaptation results on standard cross-domain digit classification tasks and a new more difficult cross-modality object classification task.
This paper describes InfoGAN, an information-theoretic extension to the Gener-ative Adversarial Network that is able to learn disentangled representations in a completely unsupervised manner. InfoGAN is a generative adversarial network that also maximizes the mutual information between a small subset of the latent variables and the observation. We derive a lower bound of the mutual information objective that can be optimized efficiently. Specifically, InfoGAN successfully disentangles writing styles from digit shapes on the MNIST dataset, pose from lighting of 3D rendered images, and background digits from the central digit on the SVHN dataset. It also discovers visual concepts that include hair styles, pres-ence/absence of eyeglasses, and emotions on the CelebA face dataset. Experiments show that InfoGAN learns interpretable representations that are competitive with representations learned by existing supervised methods.
Single modality action recognition on RGB or depth sequences has been extensively explored recently. It is generally accepted that each of these two modalities has different strengths and limitations for the task of action recognition. Therefore, analysis of the RGB+D videos can help us to better study the complementary properties of these two types of modalities and achieve higher levels of performance. In this paper, we propose a new deep autoencoder based shared-specific feature factorization network to separate input multimodal signals into a hierarchy of components. Further, based on the structure of the features, a structured sparsity learning machine is proposed which utilizes mixed norms to apply regularization within components and group selection between them for better classification performance. Our experimental results show the effectiveness of our cross-modality feature analysis framework by achieving state-of-the-art accuracy for action classification on four challenging benchmark datasets, for which we reduce the error rate by more than 40% in three datasets and saturating the benchmark with perfect accuracy for the other one.
Recognizing human actions from unknown and unseen (novel) views is a challenging problem. We propose a Robust Non-Linear Knowledge Transfer Model (R-NKTM) for human action recognition from novel views. The proposed R-NKTM is a deep fully-connected neural network that transfers knowledge of human actions from any unknown view to a shared high-level virtual view by finding a non-linear virtual path that connects the views. The R-NKTM is learned from dense trajectories of synthetic 3D human models fitted to real motion capture data and generalizes to real videos of human actions. The strength of our technique is that we learn a single R-NKTM for all actions and all viewpoints for knowledge transfer of any real human action video without the need for re-training or fine-tuning the model. Thus, R-NKTM can efficiently scale to incorporate new action classes. R-NKTM is learned with dummy labels and does not require knowledge of the camera viewpoint at any stage. Experiments on three benchmark cross-view human action datasets show that our method outperforms existing state-of-the-art.
Convolutional Neural Networks (CNNs) have been established as a powerful class of models for image recognition problems. Encouraged by these results, we provide an extensive empirical evaluation of CNNs on large-scale video classification using a new dataset of 1 million YouTube videos belonging to 487 classes. We study multiple approaches for extending the connectivity of a CNN in time domain to take advantage of local spatio-temporal information and suggest a multiresolution, foveated architecture as a promising way of speeding up the training. Our best spatio-temporal networks display significant performance improvements compared to strong feature-based baselines (55.3% to 63.9%), but only a surprisingly modest improvement compared to single-frame models (59.3% to 60.9%). We further study the generalization performance of our best model by retraining the top layers on the UCF-101 Action Recognition dataset and observe significant performance improvements compared to the UCF-101 baseline model (63.3% up from 43.9%).
Human actions in video sequences are three-dimensional (3D) spatio-temporal
signals characterizing both the visual appearance and motion dynamics of the
involved humans and objects. Inspired by the success of convolutional neural
networks (CNN) for image classification, recent attempts have been made to
learn 3D CNNs for recognizing human actions in videos. However, partly due to
the high complexity of training 3D convolution kernels and the need for large
quantities of training videos, only limited success has been reported. This has
triggered us to investigate in this paper a new deep architecture which can
handle 3D signals more effectively. Specifically, we propose factorized
spatio-temporal convolutional networks (FstCN) that factorize the original 3D
convolution kernel learning as a sequential process of learning 2D spatial
kernels in the lower layers (called spatial convolutional layers), followed by
learning 1D temporal kernels in the upper layers (called temporal convolutional
layers). We introduce a novel transformation and permutation operator to make
factorization in FstCN possible. Moreover, to address the issue of sequence
alignment, we propose an effective training and inference strategy based on
sampling multiple video clips from a given action video sequence. We have
tested FstCN on two commonly used benchmark datasets (UCF-101 and HMDB-51).
Without using auxiliary training videos to boost the performance, FstCN
outperforms existing CNN based methods and achieves comparable performance with
a recent method that benefits from using auxiliary training videos.
We report on the methods used in our recent DeepEnsembleCoco submission to
the PASCAL VOC 2012 challenge, which achieves state-of-the-art performance on
the object detection task. Our method is a variant of the R-CNN model proposed
Girshick:CVPR14 with two key improvements to training and evaluation. First,
our method constructs an ensemble of deep CNN models with different
architectures that are complementary to each other. Second, we augment the
PASCAL VOC training set with images from the Microsoft COCO dataset to
significantly enlarge the amount training data. Importantly, we select a subset
of the Microsoft COCO images to be consistent with the PASCAL VOC task. Results
on the PASCAL VOC evaluation server show that our proposed method outperform
all previous methods on the PASCAL VOC 2012 detection task at time of
submission.
Generative Adversarial Nets [8] were recently introduced as a novel way to
train generative models. In this work we introduce the conditional version of
generative adversarial nets, which can be constructed by simply feeding the
data, y, we wish to condition on to both the generator and discriminator. We
show that this model can generate MNIST digits conditioned on class labels. We
also illustrate how this model could be used to learn a multi-modal model, and
provide preliminary examples of an application to image tagging in which we
demonstrate how this approach can generate descriptive tags which are not part
of training labels.
We investigate architectures of discriminatively trained deep Convolutional
Networks (ConvNets) for action recognition in video. The challenge is to
capture the complementary information on appearance from still frames and
motion between frames. We also aim to incorporate into the network design
aspects of the best performing hand-crafted features.
Our contribution is three-fold. First, we propose a two-stream ConvNet
architecture which incorporates spatial and temporal networks. Second, we
demonstrate that a ConvNet trained on multi-frame dense optical flow is able to
achieve very good performance in spite of limited training data. Finally, we
show that multi-task learning, applied to two different action classification
datasets, can be used to increase the amount of training data and improve the
performance on both.
Our architecture is trained and evaluated on the standard video actions
benchmarks of UCF-101 and HMDB-51, where it matches the state of the art. It
also exceeds by a large margin previous attempts to use deep nets for video
classification.
Recently dense trajectories were shown to be an efficient video representation for action recognition and achieved state-of-the-art results on a variety of datasets. This paper improves their performance by taking into account camera motion to correct them. To estimate camera motion, we match feature points between frames using SURF descriptors and dense optical flow, which are shown to be complementary. These matches are, then, used to robustly estimate a homography with RANSAC. Human motion is in general different from camera motion and generates inconsistent matches. To improve the estimation, a human detector is employed to remove these matches. Given the estimated camera motion, we remove trajectories consistent with it. We also use this estimation to cancel out camera motion from the optical flow. This significantly improves motion-based descriptors, such as HOF and MBH. Experimental results on four challenging action datasets (i.e., Hollywood2, HMDB51, Olympic Sports and UCF50) significantly outperform the current state of the art.
We propose a feature, the Histogram of Oriented Normal Vectors (HONV), designed specifically to capture local geometric characteristics for object recognition with a depth sensor. Through our derivation, the normal vector orientation represented as an ordered pair of azimuthal angle and zenith angle can be easily computed from the gradients of the depth image. We form the HONV as a concatenation of local histograms of azimuthal angle and zenith angle. Since the HONV is inherently the local distribution of the tangent plane orientation of an object surface, we use it as a feature for object detection/classification tasks. The object detection experiments on the standard RGB-D dataset [1] and a self-collected Chair-D dataset show that the HONV significantly outperforms traditional features such as HOG on the depth image and HOG on the intensity image, with an improvement of 11.6% in average precision. For object classification, the HONV achieved 5.0% improvement over state-of-the-art approaches.