Generating multi-view images from a single-view input is an essential yet challenging problem. It has broad applications in vision, graphics, and robotics. Our study indicates that the widely-used generative adversarial network (GAN) may learn "incomplete" representations due to the single-pathway framework: an encoder-decoder network followed by a discriminator network. We propose CR-GAN to address this problem. In addition to the single reconstruction path, we introduce a generation sideway to maintain the completeness of the learned embedding space. The two learning pathways collaborate and compete in a parameter-sharing manner, yielding considerably improved generalization ability to "unseen" dataset. More importantly, the two-pathway framework makes it possible to combine both labeled and unlabeled data for self-supervised learning, which further enriches the embedding space for realistic generations. The experimental results prove that CR-GAN significantly outperforms state-of-the-art methods, especially when generating from "unseen" inputs in wild conditions.
We present a signal representation framework called the sparse manifold transform that combines key ideas from sparse coding, manifold learning, and slow feature analysis. It turns non-linear transformations in the primary sensory signal space into linear interpolations in a representational embedding space while maintaining approximate invertibility. The sparse manifold transform is an unsupervised and generative framework that explicitly and simultaneously models the sparse discreteness and low-dimensional manifold structure found in natural scenes. When stacked, it also models hierarchical composition. We provide a theoretical description of the transform and demonstrate properties of the learned representation on both synthetic data and natural videos.
We present a self-supervised approach to training convolutional neural networks for dense depth estimation from monocular endoscopy data without a priori modeling of anatomy or shading. Our method only requires sequential data from monocular endoscopic videos and a multi-view stereo reconstruction method, e.g. structure from motion, that supervises learning in a sparse but accurate manner. Consequently, our method requires neither manual interaction, such as scaling or labeling, nor patient CT in the training and application phases. We demonstrate the performance of our method on sinus endoscopy data from two patients and validate depth prediction quantitatively using corresponding patient CT scans where we found submillimeter residual errors.
We present a self-supervised approach using spatio-temporal signals between video frames for action recognition. A two-stream architecture is leveraged to tangle spatial and temporal representation learning. Our task is formulated as both a sequence verification and spatio-temporal alignment tasks. The former task requires motion temporal structure understanding while the latter couples the learned motion with the spatial representation. The self-supervised pre-trained weights effectiveness is validated on the action recognition task. Quantitative evaluation shows the self-supervised approach competence on three datasets: HMDB51, UCF101, and Honda driving dataset (HDD). Further investigations to boost performance and generalize validity are still required.
We propose an approach for learning category-level semantic segmentation purely from image-level classification tags indicating presence of categories. It exploits localization cues that emerge from training classification-tasked convolutional networks, to drive a "self-supervision" process that automatically labels a sparse, diverse training set of points likely to belong to classes of interest. Our approach has almost no hyperparameters, is modular, and allows for very fast training of segmentation in less than 3 minutes. It obtains competitive results on the VOC 2012 segmentation benchmark. More, significantly the modularity and fast training of our framework allows new classes to efficiently added for inference.
Manipulation of deformable objects, such as ropes and cloth, is an important but challenging problem in robotics. We present a learning-based system where a robot takes as input a sequence of images of a human manipulating a rope from an initial to goal configuration, and outputs a sequence of actions that can reproduce the human demonstration, using only monocular images as input. To perform this task, the robot learns a pixel-level inverse dynamics model of rope manipulation directly from images in a self-supervised manner, using about 60K interactions with the rope collected autonomously by the robot. The human demonstration provides a high-level plan of what to do and the low-level inverse model is used to execute the plan. We show that by combining the high and low-level plans, the robot can successfully manipulate a rope into a variety of target shapes using only a sequence of human-provided images for direction.
We propose a new self-supervised CNN pre-training technique based on a novel auxiliary task called "odd-one-out learning". In this task, the machine is asked to identify the unrelated or odd element from a set of otherwise related elements. We apply this technique to self-supervised video representation learning where we sample subsequences from videos and ask the network to learn to predict the odd video subsequence. The odd video subsequence is sampled such that it has wrong temporal order of frames while the even ones have the correct temporal order. Therefore, to generate a odd-one-out question no manual annotation is required. Our learning machine is implemented as multi-stream convolutional neural network, which is learned end-to-end. Using odd-one-out networks, we learn temporal representations for videos that generalizes to other related tasks such as action recognition. On action classification, our method obtains 60.3\% on the UCF101 dataset using only UCF101 data for training which is approximately 10% better than current state-of-the-art self-supervised learning methods. Similarly, on HMDB51 dataset we outperform self-supervised state-of-the art methods by 12.7% on action classification task.
End-to-end training from scratch of current deep architectures for new computer vision problems would require Imagenet-scale datasets, and this is not always possible. In this paper we present a method that is able to take advantage of freely available multi-modal content to train computer vision algorithms without human supervision. We put forward the idea of performing self-supervised learning of visual features by mining a large scale corpus of multi-modal (text and image) documents. We show that discriminative visual features can be learnt efficiently by training a CNN to predict the semantic context in which a particular image is more probable to appear as an illustration. For this we leverage the hidden semantic structures discovered in the text corpus with a well-known topic modeling technique. Our experiments demonstrate state of the art performance in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or natural-supervised approaches.
Many model-based Visual Odometry (VO) algorithms have been proposed in the past decade, often restricted to the type of camera optics, or the underlying motion manifold observed. We envision robots to be able to learn and perform these tasks, in a minimally supervised setting, as they gain more experience. To this end, we propose a fully trainable solution to visual ego-motion estimation for varied camera optics. We propose a visual ego-motion learning architecture that maps observed optical flow vectors to an ego-motion density estimate via a Mixture Density Network (MDN). By modeling the architecture as a Conditional Variational Autoencoder (C-VAE), our model is able to provide introspective reasoning and prediction for ego-motion induced scene-flow. Additionally, our proposed model is especially amenable to bootstrapped ego-motion learning in robots where the supervision in ego-motion estimation for a particular camera sensor can be obtained from standard navigation-based sensor fusion strategies (GPS/INS and wheel-odometry fusion). Through experiments, we show the utility of our proposed approach in enabling the concept of self-supervised learning for visual ego-motion estimation in autonomous robots.
Human parsing has recently attracted a lot of research interests due to its huge application potentials. However existing datasets have limited number of images and annotations, and lack the variety of human appearances and the coverage of challenging cases in unconstrained environment. In this paper, we introduce a new benchmark "Look into Person (LIP)" that makes a significant advance in terms of scalability, diversity and difficulty, a contribution that we feel is crucial for future developments in human-centric analysis. This comprehensive dataset contains over 50,000 elaborately annotated images with 19 semantic part labels, which are captured from a wider range of viewpoints, occlusions and background complexity. Given these rich annotations we perform detailed analyses of the leading human parsing approaches, gaining insights into the success and failures of these methods. Furthermore, in contrast to the existing efforts on improving the feature discriminative capability, we solve human parsing by exploring a novel self-supervised structure-sensitive learning approach, which imposes human pose structures into parsing results without resorting to extra supervision (i.e., no need for specifically labeling human joints in model training). Our self-supervised learning framework can be injected into any advanced neural networks to help incorporate rich high-level knowledge regarding human joints from a global perspective and improve the parsing results. Extensive evaluations on our LIP and the public PASCAL-Person-Part dataset demonstrate the superiority of our method.
A significant proportion of patients scanned in a clinical setting have follow-up scans. We show in this work that such longitudinal scans alone can be used as a form of 'free' self-supervision for training a deep network. We demonstrate this self-supervised learning for the case of T2-weighted sagittal lumbar Magnetic Resonance Images (MRIs). A Siamese convolutional neural network (CNN) is trained using two losses: (i) a contrastive loss on whether the scan is of the same person (i.e. longitudinal) or not, together with (ii) a classification loss on predicting the level of vertebral bodies. The performance of this pre-trained network is then assessed on a grading classification task. We experiment on a dataset of 1016 subjects, 423 possessing follow-up scans, with the end goal of learning the disc degeneration radiological gradings attached to the intervertebral discs. We show that the performance of the pre-trained CNN on the supervised classification task is (i) superior to that of a network trained from scratch; and (ii) requires far fewer annotated training samples to reach an equivalent performance to that of the network trained from scratch.
Progress has been achieved recently in object detection given advancements in deep learning. Nevertheless, such tools typically require a large amount of training data and significant manual effort to label objects. This limits their applicability in robotics, where solutions must scale to a large number of objects and variety of conditions. This work proposes an autonomous process for training a Convolutional Neural Network (CNN) for object detection and pose estimation in robotic setups. The focus is on detecting objects placed in cluttered, tight environments, such as a shelf with multiple objects. In particular, given access to 3D object models, several aspects of the environment are physically simulated. The models are placed in physically realistic poses with respect to their environment to generate a labeled synthetic dataset. To further improve object detection, the network self-trains over real images that are labeled using a robust multi-view pose estimation process. The proposed training process is evaluated on several existing datasets and on a dataset collected for this paper with a Motoman robotic arm. Results show that the proposed approach outperforms popular training processes relying on synthetic - but not physically realistic - data and manual annotation. The key contributions are the incorporation of physical reasoning in the synthetic data generation process and the automation of the annotation process over real images.
Intrinsically motivated spontaneous exploration is a key enabler of autonomous developmental learning in human children. It enables the discovery of skill repertoires through autotelic learning, i.e. the self-generation, self-selection, self-ordering and self-experimentation of learning goals. We present an algorithmic approach called Intrinsically Motivated Goal Exploration Processes (IMGEP) to enable similar properties of autonomous learning in machines. The IMGEP architecture relies on several principles: 1) self-generation of goals, generalized as parameterized fitness functions; 2) selection of goals based on intrinsic rewards; 3) exploration with incremental goal-parameterized policy search and exploitation with a batch learning algorithm; 4) systematic reuse of information acquired when targeting a goal for improving towards other goals. We present a particularly efficient form of IMGEP, called AMB, that uses a population-based policy and an object-centered spatio-temporal modularity. We provide several implementations of this architecture and demonstrate their ability to automatically generate a learning curriculum within several experimental setups. One of these experiments includes a real humanoid robot exploring multiple spaces of goals with several hundred continuous dimensions and with distractors. While no particular target goal is provided to these autotelic agents, this curriculum allows the discovery of diverse skills that act as stepping stones for learning more complex skills, e.g. nested tool use.
Human pose analysis is presently dominated by deep convolutional networks trained with extensive manual annotations of joint locations and beyond. To avoid the need for expensive labeling, we exploit spatiotemporal relations in training videos for self-supervised learning of pose embeddings. The key idea is to combine temporal ordering and spatial placement estimation as auxiliary tasks for learning pose similarities in a Siamese convolutional network. Since the self-supervised sampling of both tasks from natural videos can result in ambiguous and incorrect training labels, our method employs a curriculum learning idea that starts training with the most reliable data samples and gradually increases the difficulty. To further refine the training process we mine repetitive poses in individual videos which provide reliable labels while removing inconsistencies. Our pose embeddings capture visual characteristics of human pose that can boost existing supervised representations in human pose estimation and retrieval. We report quantitative and qualitative results on these tasks in Olympic Sports, Leeds Pose Sports and MPII Human Pose datasets.
This paper proposes a weakly- and self-supervised deep convolutional neural network (WSSDCNN) for content-aware image retargeting. Our network takes a source image and a target aspect ratio, and then directly outputs a retargeted image. Retargeting is performed through a shift map, which is a pixel-wise mapping from the source to the target grid. Our method implicitly learns an attention map, which leads to a content-aware shift map for image retargeting. As a result, discriminative parts in an image are preserved, while background regions are adjusted seamlessly. In the training phase, pairs of an image and its image-level annotation are used to compute content and structure losses. We demonstrate the effectiveness of our proposed method for a retargeting application with insightful analyses.
Learning visual representations with self-supervised learning has become popular in computer vision. The idea is to design auxiliary tasks where labels are free to obtain. Most of these tasks end up providing data to learn specific kinds of invariance useful for recognition. In this paper, we propose to exploit different self-supervised approaches to learn representations invariant to (i) inter-instance variations (two objects in the same class should have similar features) and (ii) intra-instance variations (viewpoint, pose, deformations, illumination, etc). Instead of combining two approaches with multi-task learning, we argue to organize and reason the data with multiple variations. Specifically, we propose to generate a graph with millions of objects mined from hundreds of thousands of videos. The objects are connected by two types of edges which correspond to two types of invariance: "different instances but a similar viewpoint and category" and "different viewpoints of the same instance". By applying simple transitivity on the graph with these edges, we can obtain pairs of images exhibiting richer visual invariance. We use this data to train a Triplet-Siamese network with VGG16 as the base architecture and apply the learned representations to different recognition tasks. For object detection, we achieve 63.2% mAP on PASCAL VOC 2007 using Fast R-CNN (compare to 67.3% with ImageNet pre-training). For the challenging COCO dataset, our method is surprisingly close (23.5%) to the ImageNet-supervised counterpart (24.4%) using the Faster R-CNN framework. We also show that our network can perform significantly better than the ImageNet network in the surface normal estimation task.
Exiting deep-learning based dense stereo matching methods often rely on ground-truth disparity maps as the training signals, which are however not always available in many situations. In this paper, we design a simple convolutional neural network architecture that is able to learn to compute dense disparity maps directly from the stereo inputs. Training is performed in an end-to-end fashion without the need of ground-truth disparity maps. The idea is to use image warping error (instead of disparity-map residuals) as the loss function to drive the learning process, aiming to find a depth-map that minimizes the warping error. While this is a simple concept well-known in stereo matching, to make it work in a deep-learning framework, many non-trivial challenges must be overcome, and in this work we provide effective solutions. Our network is self-adaptive to different unseen imageries as well as to different camera settings. Experiments on KITTI and Middlebury stereo benchmark datasets show that our method outperforms many state-of-the-art stereo matching methods with a margin, and at the same time significantly faster.
Self-supervised learning (SSL) is a reliable learning mechanism in which a robot enhances its perceptual capabilities. Typically, in SSL a trusted, primary sensor cue provides supervised training data to a secondary sensor cue. In this article, a theoretical analysis is performed on the fusion of the primary and secondary cue in a minimal model of SSL. A proof is provided that determines the specific conditions under which it is favorable to perform fusion. In short, it is favorable when (i) the prior on the target value is strong or (ii) the secondary cue is sufficiently accurate. The theoretical findings are validated with computational experiments. Subsequently, a real-world case study is performed to investigate if fusion in SSL is also beneficial when assumptions of the minimal model are not met. In particular, a flying robot learns to map pressure measurements to sonar height measurements and then fuses the two, resulting in better height estimation. Fusion is also beneficial in the opposite case, when pressure is the primary cue. The analysis and results are encouraging to study SSL fusion also for other robots and sensors.
Generating music medleys is about finding an optimal permutation of a given set of music clips. Toward this goal, we propose a self-supervised learning task, called the music puzzle game, to train neural network models to learn the sequential patterns in music. In essence, such a game requires machines to correctly sort a few multisecond music fragments. In the training stage, we learn the model by sampling multiple non-overlapping fragment pairs from the same songs and seeking to predict whether a given pair is consecutive and is in the correct chronological order. For testing, we design a number of puzzle games with different difficulty levels, the most difficult one being music medley, which requiring sorting fragments from different songs. On the basis of state-of-the-art Siamese convolutional network, we propose an improved architecture that learns to embed frame-level similarity scores computed from the input fragment pairs to a common space, where fragment pairs in the correct order can be more easily identified. Our result shows that the resulting model, dubbed as the similarity embedding network (SEN), performs better than competing models across different games, including music jigsaw puzzle, music sequencing, and music medley. Example results can be found at our project website, https://remyhuang.github.io/DJnet.
Current state-of-the-art solutions for motion capture from a single camera are optimization driven: they optimize the parameters of a 3D human model so that its re-projection matches measurements in the video (e.g. person segmentation, optical flow, keypoint detections etc.). Optimization models are susceptible to local minima. This has been the bottleneck that forced using clean green-screen like backgrounds at capture time, manual initialization, or switching to multiple cameras as input resource. In this work, we propose a learning based motion capture model for single camera input. Instead of optimizing mesh and skeleton parameters directly, our model optimizes neural network weights that predict 3D shape and skeleton configurations given a monocular RGB video. Our model is trained using a combination of strong supervision from synthetic data, and self-supervision from differentiable rendering of (a) skeletal keypoints, (b) dense 3D mesh motion, and (c) human-background segmentation, in an end-to-end framework. Empirically we show our model combines the best of both worlds of supervised learning and test-time optimization: supervised learning initializes the model parameters in the right regime, ensuring good pose and surface initialization at test time, without manual effort. Self-supervision by back-propagating through differentiable rendering allows (unsupervised) adaptation of the model to the test data, and offers much tighter fit than a pretrained fixed model. We show that the proposed model improves with experience and converges to low-error solutions where previous optimization methods fail.