Convolutional neural networks (CNNs) can model complicated non-linear relations between images. However, they are notoriously sensitive to small changes in the input. Most CNNs trained to describe image-to-image mappings generate temporally unstable results when applied to video sequences, leading to flickering artifacts and other inconsistencies over time. In order to use CNNs for video material, previous methods have relied on estimating dense frame-to-frame motion information (optical flow) in the training and/or the inference phase, or by exploring recurrent learning structures. We take a different approach to the problem, posing temporal stability as a regularization of the cost function. The regularization is formulated to account for different types of motion that can occur between frames, so that temporally stable CNNs can be trained without the need for video material or expensive motion estimation. The training can be performed as a fine-tuning operation, without architectural modifications of the CNN. Our evaluation shows that the training strategy leads to large improvements in temporal smoothness. Moreover, in situations where the quantity of training data is limited, the regularization can help in boosting the generalization performance to a much larger extent than what is possible with na\"ive augmentation strategies.
We propose a 3D object detection method for autonomous driving by fully exploiting the sparse and dense, semantic and geometry information in stereo imagery. Our method, called Stereo R-CNN, extends Faster R-CNN for stereo inputs to simultaneously detect and associate object in left and right images. We add extra branches after stereo Region Proposal Network (RPN) to predict sparse keypoints, viewpoints, and object dimensions, which are combined with 2D left-right boxes to calculate a coarse 3D object bounding box. We then recover the accurate 3D bounding box by a region-based photometric alignment using left and right RoIs. Our method does not require depth input and 3D position supervision, however, outperforms all existing fully supervised image-based methods. Experiments on the challenging KITTI dataset show that our method outperforms the state-of-the-art stereo-based method by around 30% AP on both 3D detection and 3D localization tasks. Code will be made publicly available.
Biomedical image segmentation is an important task in many medical applications. Segmentation methods based on convolutional neural networks attain state-of-the-art accuracy; however, they typically rely on supervised training with large labeled datasets. Labeling datasets of medical images requires significant expertise and time, and is infeasible at large scales. To tackle the lack of labeled data, researchers use techniques such as hand-engineered preprocessing steps, hand-tuned architectures, and data augmentation. However, these techniques involve costly engineering efforts, and are typically dataset-specific. We present an automated data augmentation method for medical images. We demonstrate our method on the task of segmenting magnetic resonance imaging (MRI) brain scans, focusing on the one-shot segmentation scenario -- a practical challenge in many medical applications. Our method requires only a single segmented scan, and leverages other unlabeled scans in a semi-supervised approach. We learn a model of transforms from the images, and use the model along with the labeled example to synthesize additional labeled training examples for supervised segmentation. Each transform is comprised of a spatial deformation field and an intensity change, enabling the synthesis of complex effects such as variations in anatomy and image acquisition procedures. Augmenting the training of a supervised segmenter with these new examples provides significant improvements over state-of-the-art methods for one-shot biomedical image segmentation. Our code is available at https://github.com/xamyzhao/brainstorm.
This paper focuses on Bayesian Optimization - typically considered with continuous inputs - for discrete search input spaces, including integer, categorical or graph structured input variables. In Gaussian process-based Bayesian Optimization a problem arises, as it is not straightforward to define a proper kernel on discrete input structures, where no natural notion of smoothness or similarity could be provided. We propose COMBO, a method that represents values of discrete variables as vertices of a graph and then use the diffusion kernel on that graph. As the graph size explodes with the number of categorical variables and categories, we propose the graph Cartesian product to decompose the graph into smaller sub-graphs, enabling kernel computation in linear time with respect to the number of input variables. Moreover, in our formulation we learn a scale parameter per subgraph. In empirical studies on four discrete optimization problems we demonstrate that our method is on par or outperforms the state-of-the-art in discrete Bayesian optimization.
Recent works have highlighted the strengths of the Transformer architecture for dealing with sequence tasks. At the same time, neural architecture search has advanced to the point where it can outperform human-designed models. The goal of this work is to use architecture search to find a better Transformer architecture. We first construct a large search space inspired by the recent advances in feed-forward sequential models and then run evolutionary architecture search, seeding our initial population with the Transformer. To effectively run this search on the computationally expensive WMT 2014 English-German translation task, we develop the progressive dynamic hurdles method, which allows us to dynamically allocate more resources to more promising candidate models. The architecture found in our experiments - the Evolved Transformer - demonstrates consistent improvement over the Transformer on four well-established language tasks: WMT 2014 English-German, WMT 2014 English-French, WMT 2014 English-Czech and LM1B. At big model size, the Evolved Transformer is twice as efficient as the Transformer in FLOPS without loss in quality. At a much smaller - mobile-friendly - model size of ~7M parameters, the Evolved Transformer outperforms the Transformer by 0.7 BLEU on WMT'14 English-German.
Machine-learning models have demonstrated great success in learning complex patterns that enable them to make predictions about unobserved data. In addition to using models for prediction, the ability to interpret what a model has learned is receiving an increasing amount of attention. However, this increased focus has led to considerable confusion about the notion of interpretability. In particular, it is unclear how the wide array of proposed interpretation methods are related, and what common concepts can be used to evaluate them. We aim to address these concerns by defining interpretability in the context of machine learning and introducing the Predictive, Descriptive, Relevant (PDR) framework for discussing interpretations. The PDR framework provides three overarching desiderata for evaluation: predictive accuracy, descriptive accuracy and relevancy, with relevancy judged relative to a human audience. Moreover, to help manage the deluge of interpretation methods, we introduce a categorization of existing techniques into model-based and post-hoc categories, with sub-groups including sparsity, modularity and simulatability. To demonstrate how practitioners can use the PDR framework to evaluate and understand interpretations, we provide numerous real-world examples. These examples highlight the often under-appreciated role played by human audiences in discussions of interpretability. Finally, based on our framework, we discuss limitations of existing methods and directions for future work. We hope that this work will provide a common vocabulary that will make it easier for both practitioners and researchers to discuss and choose from the full range of interpretation methods.
Modern inexpensive imaging sensors suffer from inherent hardware constraints which often result in captured images of poor quality. Among the most common ways to deal with such limitations is to rely on burst photography, which nowadays acts as the backbone of all modern smartphone imaging applications. In this work, we focus on the fact that every frame of a burst sequence can be accurately described by a forward (physical) model. This in turn allows us to restore a single image of higher quality from a sequence of low quality images as the solution of an optimization problem. Inspired by an extension of the gradient descent method that can handle non-smooth functions, namely the proximal gradient descent, and modern deep learning techniques, we propose a convolutional iterative network with a transparent architecture. Our network, uses a burst of low quality image frames and is able to produce an output of higher image quality recovering fine details which are not distinguishable in any of the original burst frames. We focus both on the burst photography pipeline as a whole, i.e. burst demosaicking and denoising, as well as on the traditional Gaussian denoising task. The developed method demonstrates consistent state-of-the art performance across the two tasks and as opposed to other recent deep learning approaches does not have any inherent restrictions either to the number of frames or their ordering.
The Super-Resolution Generative Adversarial Network (SRGAN) is a seminal work that is capable of generating realistic textures during single image super-resolution. However, the hallucinated details are often accompanied with unpleasant artifacts. To further enhance the visual quality, we thoroughly study three key components of SRGAN - network architecture, adversarial loss and perceptual loss, and improve each of them to derive an Enhanced SRGAN (ESRGAN). In particular, we introduce the Residual-in-Residual Dense Block (RRDB) without batch normalization as the basic network building unit. Moreover, we borrow the idea from relativistic GAN to let the discriminator predict relative realness instead of the absolute value. Finally, we improve the perceptual loss by using the features before activation, which could provide stronger supervision for brightness consistency and texture recovery. Benefiting from these improvements, the proposed ESRGAN achieves consistently better visual quality with more realistic and natural textures than SRGAN and won the first place in the PIRM2018-SR Challenge. The code is available at https://github.com/xinntao/ESRGAN .
In recent years, deep neural models have been widely adopted for text matching tasks, such as question answering and information retrieval, showing improved performance as compared with previous methods. In this paper, we introduce the MatchZoo toolkit that aims to facilitate the designing, comparing and sharing of deep text matching models. Specifically, the toolkit provides a unified data preparation module for different text matching problems, a flexible layer-based model construction process, and a variety of training objectives and evaluation metrics. In addition, the toolkit has implemented two schools of representative deep text matching models, namely representation-focused models and interaction-focused models. Finally, users can easily modify existing models, create and share their own models for text matching in MatchZoo.