CVPR 2019 Papers

We consider the problem of providing dense segmentation masks for object discovery in videos. We formulate the object discovery problem as foreground motion clustering, where the goal is to cluster foreground pixels in videos into different objects. We introduce a novel pixel-trajectory recurrent neural network that learns feature embeddings of foreground pixel trajectories linked across time. By clustering the pixel trajectories using the learned feature embeddings, our method establishes correspondences between foreground object masks across video frames. To demonstrate the effectiveness of our framework for object discovery, we conduct experiments on commonly used datasets for motion segmentation, where we achieve state-of-the-art performance.

Multi-class and multi-scale object detection for autonomous driving is challenging because of the high variation in object scales and the cluttered background in complex street scenes. Context information and high-resolution features are the keys to achieve a good performance in multi-scale object detection. However, context information is typically unevenly distributed, and the high-resolution feature map also contains distractive low-level features. In this paper, we propose a location-aware deformable convolution and a backward attention filtering to improve the detection performance. The location-aware deformable convolution extracts the unevenly distributed context features by sampling the input from where informative context exists. Different from the original deformable convolution, the proposed method applies an individual convolutional layer on each input sampling grid location to obtain a wide and unique receptive field for a better offset estimation. Meanwhile, the backward attention filtering module filters the high-resolution feature map by highlighting the informative features and suppressing the distractive features using the semantic features from the deep layers. Extensive experiments are conducted on the KITTI object detection and PASCAL VOC 2007 datasets. The proposed method shows an average 6% performance improvement over the Faster R-CNN baseline, and it has the top-3 performance on the KITTI leaderboard with the fastest processing speed.

Common object counting in a natural scene is a challenging problem in computer vision with numerous real-world applications. Existing image-level supervised common object counting approaches only predict the global object count and rely on additional instance-level supervision to also determine object locations. We propose an image-level supervised approach that provides both the global object count and the spatial distribution of object instances by constructing an object category density map. Motivated by psychological studies, we further reduce image-level supervision using a limited object count information (up to four). To the best of our knowledge, we are the first to propose image-level supervised density map estimation for common object counting and demonstrate its effectiveness in image-level supervised instance segmentation. Comprehensive experiments are performed on the PASCAL VOC and COCO datasets. Our approach outperforms existing methods, including those using instance-level supervision, on both datasets for common object counting. Moreover, our approach improves state-of-the-art image-level supervised instance segmentation with a relative gain of 17.8% in terms of average best overlap, on the PASCAL VOC 2012 dataset.

Visual Question Answering (VQA) in its ideal form lets us study reasoning in the joint space of vision and language and serves as a proxy for the AI task of scene understanding. However, most VQA benchmarks to date are focused on questions such as simple counting, visual attributes, and object detection that do not require reasoning or knowledge beyond what is in the image. In this paper, we address the task of knowledge-based visual question answering and provide a benchmark, called OK-VQA, where the image content is not sufficient to answer the questions, encouraging methods that rely on external knowledge resources. Our new dataset includes more than 14,000 questions that require external knowledge to answer. We show that the performance of the state-of-the-art VQA models degrades drastically in this new setting. Our analysis shows that our knowledge-based VQA task is diverse, difficult, and large compared to previous knowledge-based VQA datasets. We hope that this dataset enables researchers to open up new avenues for research in this domain.

Channel pruning can significantly accelerate and compress deep neural networks. Many channel pruning works utilize structured sparsity regularization to zero out all the weights in some channels and automatically obtain structure-sparse network in training stage. However, these methods apply structured sparsity regularization on each layer separately where the correlations between consecutive layers are omitted. In this paper, we first combine one out-channel in current layer and the corresponding in-channel in next layer as a regularization group, namely out-in-channel. Our proposed Out-In-Channel Sparsity Regularization (OICSR) considers correlations between successive layers to further retain predictive power of the compact network. Training with OICSR thoroughly transfers discriminative features into a fraction of out-in-channels. Correspondingly, OICSR measures channel importance based on statistics computed from two consecutive layers, not individual layer. Finally, a global greedy pruning algorithm is designed to remove redundant out-in-channels in an iterative way. Our method is comprehensively evaluated with various CNN architectures including CifarNet, AlexNet, ResNet, DenseNet and PreActSeNet on CIFAR-10, CIFAR-100 and ImageNet-1K datasets. Notably, on ImageNet-1K, we reduce 37.2% FLOPs on ResNet-50 while outperforming the original model by 0.22% top-1 accuracy.

Single image super-resolution, as a high dimensional structured prediction problem, aims to characterize fine-grain information given a low-resolution sample. Recent advances in convolutional neural networks are introduced into super-resolution and push forward progress in this field. Current studies have achieved impressive performance by manually designing deep residual neural networks but overly relies on practical experience. In this paper, we propose to adopt an ordinary differential equation (ODE)-inspired design scheme for single image super-resolution, which have brought us a new understanding of ResNet in classification problems. Not only is it interpretable for super-resolution but it provides a reliable guideline on network designs. By casting the numerical schemes in ODE as blueprints, we derive two types of network structures: LF-block and RK-block, which correspond to the Leapfrog method and Runge-Kutta method in numerical ordinary differential equations. We evaluate our models on benchmark datasets, and the results show that our methods surpass the state-of-the-arts while keeping comparable parameters and operations.

We present a novel model called OCGAN for the classical problem of one-class novelty detection, where, given a set of examples from a particular class, the goal is to determine if a query example is from the same class. Our solution is based on learning latent representations of in-class examples using a de-noising auto-encoder network. The key contribution of our work is our proposal to explicitly constrain the latent space to exclusively represent the given class. In order to accomplish this goal, firstly, we force the latent space to have bounded support by introducing a tanh activation in the encoder's output layer. Secondly, using a discriminator in the latent space that is trained adversarially, we ensure that encoded representations of in-class examples resemble uniform random samples drawn from the same bounded space. Thirdly, using a second adversarial discriminator in the input space, we ensure all randomly drawn latent samples generate examples that look real. Finally, we introduce a gradient-descent based sampling technique that explores points in the latent space that generate potential out-of-class examples, which are fed back to the network to further train it to generate in-class examples from those points. The effectiveness of the proposed method is measured across four publicly available datasets using two one-class novelty detection protocols where we achieve state-of-the-art results.

Importance of visual context in scene understanding tasks is well recognized in the computer vision community. However, to what extent the computer vision models are dependent on the context to make their predictions is unclear. A model overly relying on context will fail when encountering objects in different contexts than in training data and hence it is important to identify these dependencies before we can deploy the models in the real-world. We propose a method to quantify the sensitivity of black-box vision models to visual context by editing images to remove selected objects and measuring the response of the target models. We apply this methodology on two tasks, image classification and semantic segmentation, and discover undesirable dependency between objects and context, for example that "sidewalk" segmentation is very sensitive to the presence of "cars" in the image. We propose an object removal based data augmentation solution to mitigate this dependency and increase the robustness of classification and segmentation models to contextual variations. Our experiments show that the proposed data augmentation helps these models improve the performance in out-of-context scenarios, while preserving the performance on regular data.

We invest the problem of weakly-supervised video grounding, where only video-level sentences are provided. This is a challenging task, and previous Multi-Instance Learning (MIL) based image grounding methods turn to fail in the video domain. Recent work attempts to decompose the video-level MIL into frame-level MIL by applying weighted sentence-frame ranking loss over frames, but it is not robust and does not exploit the rich temporal information in videos. In this work, we address these issues by extending frame-level MIL with a false positive frame-bag constraint and modeling the visual feature consistency in the video. In specific, we design a contextual similarity between semantic and visual features to deal with sparse objects association across frames. Furthermore, we leverage temporal coherence by strengthening the clustering effect of similar features in the visual space. We conduct an extensive evaluation on YouCookII and RoboWatch datasets, and demonstrate our method significantly outperforms prior state-of-the-art methods.

Generating diverse yet specific data is the goal of the generative adversarial network (GAN), but it suffers from the problem of mode collapse. We introduce the concept of normalized diversity which force the model to preserve the normalized pairwise distance between the sparse samples from a latent parametric distribution and their corresponding high-dimensional outputs. The normalized diversification aims to unfold the manifold of unknown topology and non-uniform distribution, which leads to safe interpolation between valid latent variables. By alternating the maximization over the pairwise distance and updating the total distance (normalizer), we encourage the model to actively explore in the high-dimensional output space. We demonstrate that by combining the normalized diversity loss and the adversarial loss, we generate diverse data without suffering from mode collapsing. Experimental results show that our method achieves consistent improvement on unsupervised image generation, conditional image generation and hand pose estimation over strong baselines.

Unconditional image generation has recently been dominated by generative adversarial networks (GANs). GAN methods train a generator which regresses images from random noise vectors, as well as a discriminator that attempts to differentiate between the generated images and a training set of real images. GANs have shown amazing results at generating realistic looking images. Despite their success, GANs suffer from critical drawbacks including: unstable training and mode-dropping. The weaknesses in GANs have motivated research into alternatives including: variational auto-encoders (VAEs), latent embedding learning methods (e.g. GLO) and nearest-neighbor based implicit maximum likelihood estimation (IMLE). Unfortunately at the moment, GANs still significantly outperform the alternative methods for image generation. In this work, we present a novel method - Generative Latent Nearest Neighbors (GLANN) - for training generative models without adversarial training. GLANN combines the strengths of IMLE and GLO in a way that overcomes the main drawbacks of each method. Consequently, GLANN generates images that are far better than GLO and IMLE. Our method does not suffer from mode collapse which plagues GAN training and is much more stable. Qualitative results show that GLANN outperforms a baseline consisting of 800 GANs and VAEs on commonly used datasets. Our models are also shown to be effective for training truly non-adversarial unsupervised image translation.

The field of image denoising is currently dominated by discriminative deep learning methods that are trained on pairs of noisy input and clean target images. Recently it has been shown that such methods can also be trained without clean targets. Instead, independent pairs of noisy images can be used, in an approach known as Noise2Noise (N2N). Here, we introduce Noise2Void (N2V), a training scheme that takes this idea one step further. It does not require noisy image pairs, nor clean target images. Consequently, N2V allows us to train directly on the body of data to be denoised and can therefore be applied when other methods cannot. Especially interesting is the application to biomedical image data, where the acquisition of training targets, clean or noisy, is frequently not possible. We compare the performance of N2V to approaches that have either clean target images and/or noisy image pairs available. Intuitively, N2V cannot be expected to outperform methods that have more information available during training. Still, we observe that the denoising performance of Noise2Void drops in moderation and compares favorably to training-free denoising methods.

Benefit from large-scale training datasets, deep Convolutional Neural Networks(CNNs) have achieved impressive results in face recognition(FR). However, tremendous scale of datasets inevitably lead to noisy data, which obviously reduce the performance of the trained CNN models. Kicking out wrong labels from large-scale FR datasets is still very expensive, although some cleaning approaches are proposed. According to the analysis of the whole process of training CNN models supervised by angular margin based loss(AM-Loss) functions, we find that the distribution of training samples implicitly reflects their probability of being clean. Thus, we propose a novel training paradigm that employs the idea of weighting samples based on the above probability. Without any prior knowledge of noise, we can train high performance CNN models with largescale FR datasets. Experiments demonstrate the effectiveness of our training paradigm. The codes are available at https://github.com/huangyangyu/NoiseFace.

In this paper, we present LidarStereoNet, the first unsupervised Lidar-stereo fusion network, which can be trained in an end-to-end manner without the need of ground truth depth maps. By introducing a novel "Feedback Loop" to connect the network input with output, LidarStereoNet could tackle both noisy Lidar points and misalignment between sensors that have been ignored in existing Lidar-stereo fusion work. Besides, we propose to incorporate the piecewise planar model into the network learning to further constrain depths to conform to the underlying 3D geometry. Extensive quantitative and qualitative evaluations on both real and synthetic datasets demonstrate the superiority of our method, which outperforms state-of-the-art stereo matching, depth completion and Lidar-Stereo fusion approaches significantly.

Generalizing eye tracking to new subjects/environments remains challenging for existing appearance-based methods. To address this issue, we propose to leverage on eye movement dynamics inspired by neurological studies. Studies show that there exist several common eye movement types, independent of viewing contents and subjects, such as fixation, saccade, and smooth pursuits. Incorporating generic eye movement dynamics can therefore improve the generalization capabilities. In particular, we propose a novel Dynamic Gaze Transition Network (DGTN) to capture the underlying eye movement dynamics and serve as the topdown gaze prior. Combined with the bottom-up gaze measurements from the deep convolutional neural network, our method achieves better performance for both within-dataset and cross-dataset evaluations compared to state-of-the-art. In addition, a new DynamicGaze dataset is also constructed to study eye movement dynamics and eye gaze estimation.

We propose an end-to-end approach for phrase grounding in images. Unlike prior methods that typically attempt to ground each phrase independently by building an image-text embedding, our architecture formulates grounding of multiple phrases as a sequential and contextual process. Specifically, we encode region proposals and all phrases into two stacks of LSTM cells, along with so-far grounded phrase-region pairs. These LSTM stacks collectively capture context for grounding of the next phrase. The resulting architecture, which we call SeqGROUND, supports many-to-many matching by allowing an image region to be matched to multiple phrases and vice versa. We show competitive performance on the Flickr30K benchmark dataset and, through ablation studies, validate the efficacy of sequential grounding as well as individual design choices in our model architecture.

Learning general image representations has proven key to the success of many computer vision tasks. For example, many approaches to image understanding problems rely on deep networks that were initially trained on ImageNet, mostly because the learned features are a valuable starting point to learn from limited labeled data. However, when it comes to 3D motion capture of multiple people, these features are only of limited use. In this paper, we therefore propose an approach to learning features that are useful for this purpose. To this end, we introduce a self-supervised approach to learning what we call a neural scene decomposition (NSD) that can be exploited for 3D pose estimation. NSD comprises three layers of abstraction to represent human subjects: spatial layout in terms of bounding-boxes and relative depth; a 2D shape representation in terms of an instance segmentation mask; and subject-specific appearance and 3D pose information. By exploiting self-supervision coming from multiview data, our NSD model can be trained end-to-end without any 2D or 3D supervision. In contrast to previous approaches, it works for multiple persons and full-frame images. Because it encodes 3D geometry, NSD can then be effectively leveraged to train a 3D pose estimation network from small amounts of annotated data.

We introduce an end-to-end deep-learning framework for 3D medical image registration. In contrast to existing approaches, our framework combines two registration methods: an affine registration and a vector momentum-parameterized stationary velocity field (vSVF) model. Specifically, it consists of three stages. In the first stage, a multi-step affine network predicts affine transform parameters. In the second stage, we use a U-Net-like network to generate a momentum, from which a velocity field can be computed via smoothing. Finally, in the third stage, we employ a self-iterable map-based vSVF component to provide a non-parametric refinement based on the current estimate of the transformation map. Once the model is trained, a registration is completed in one forward pass. To evaluate the performance, we conducted longitudinal and cross-subject experiments on 3D magnetic resonance images (MRI) of the knee of the Osteoarthritis Initiative (OAI) dataset. Results show that our framework achieves comparable performance to state-of-the-art medical image registration approaches, but it is much faster, with a better control of transformation regularity including the ability to produce approximately symmetric transformations, and combining affine as well as non-parametric registration.

Real-world applications of object recognition often require the solution of multiple tasks in a single platform. Under the standard paradigm of network fine-tuning, an entirely new CNN is learned per task, and the final network size is independent of task complexity. This is wasteful, since simple tasks require smaller networks than more complex tasks, and limits the number of tasks that can be solved simultaneously. To address these problems, we propose a transfer learning procedure, denoted NetTailor, in which layers of a pre-trained CNN are used as universal blocks that can be combined with small task-specific layers to generate new networks. Besides minimizing classification error, the new network is trained to mimic the internal activations of a strong unconstrained CNN, and minimize its complexity by the combination of 1) a soft-attention mechanism over blocks and 2) complexity regularization constraints. In this way, NetTailor can adapt the network architecture, not just its weights, to the target task. Experiments show that networks adapted to simple tasks, such as character or traffic sign recognition, become significantly smaller than those adapted to hard tasks, such as fine-grained recognition. More importantly, due to the modular nature of the procedure, this reduction in network complexity is achieved without compromise of either parameter sharing across tasks, or classification accuracy.