CVPR 2017 Papers

Due to the importance of zero-shot learning, the number of proposed approaches has increased steadily recently. We argue that it is time to take a step back and to analyze the status quo of the area. The purpose of this paper is three-fold. First, given the fact that there is no agreed upon zero-shot learning benchmark, we first define a new benchmark by unifying both the evaluation protocols and data splits. This is an important contribution as published results are often not comparable and sometimes even flawed due to, e.g. pre-training on zero-shot test classes. Second, we compare and analyze a significant number of the state-of-the-art methods in depth, both in the classic zero-shot setting but also in the more realistic generalized zero-shot setting. Finally, we discuss limitations of the current status of the area which can be taken as a basis for advancing it.

Zero-shot learning, a special case of unsupervised domain adaptation where the source and target domains have disjoint label spaces, has become increasingly popular in the computer vision community. In this paper, we propose a novel zero-shot learning method based on discriminative sparse non-negative matrix factorization. The proposed approach aims to identify a set of common high-level semantic components across the two domains via non-negative sparse matrix factorization, while enforcing the representation vectors of the images in this common component-based space to be discriminatively aligned with the attribute-based label representation vectors. To fully exploit the aligned semantic information contained in the learned representation vectors of the instances, we develop a label propagation based testing procedure to classify the unlabeled instances from the unseen classes in the target domain. We conduct experiments on four standard zero-shot learning image datasets, by comparing the proposed approach to the state-of-the-art zero-shot learning methods. The empirical results demonstrate the efficacy of the proposed approach.

Recently, zero-shot action recognition (ZSAR) has emerged with the explosive growth of action categories. In this paper, we explore ZSAR from a novel perspective by adopting the Error-Correcting Output Codes (dubbed ZSECOC). Our ZSECOC equips the conventional ECOC with the additional capability of ZSAR, by addressing the domain shift problem. In particular, we learn discriminative ZSECOC for seen categories from both category-level semantics and intrinsic data structures. This procedure deals with domain shift implicitly by transferring the well-established correlations among seen categories to unseen ones. Moreover, a simple semantic transfer strategy is developed for explicitly transforming the learned embeddings of seen categories to better fit the underlying structure of unseen categories. As a consequence, our ZSECOC inherits the promising characteristics from ECOC as well as overcomes domain shift, making it more discriminative for ZSAR. We systematically evaluate ZSECOC on three realistic action benchmarks, i.e. Olympic Sports, HMDB51 and UCF101. The experimental results clearly show the superiority of ZSECOC over the state-of-the-art methods.

We address zero-shot learning using a new manifold alignment framework based on a localized multi-scale transform on graphs. Our inference approach includes a smoothness criterion for a function mapping nodes on a graph (visual representation) onto a linear space (semantic representation), which we optimize using multi-scale graph wavelets. The robustness of the ensuing scheme allows us to operate with automatically generated semantic annotations, resulting in an algorithm that is entirely free of manual supervision, and yet improves the state-of-the-art as measured on benchmark datasets.

We introduce a new large-scale data set of video URLs with densely-sampled object bounding box annotations called YouTube-BoundingBoxes (YT-BB). The data set consists of approximately 380,000 video segments about 19s long, automatically selected to feature objects in natural settings without editing or post-processing, with a recording quality often akin to that of a hand-held cell phone camera. The objects represent a subset of the COCO label set. All video segments were human-annotated with high-precision classification labels and bounding boxes at 1 frame per second. The use of a cascade of increasingly precise human annotations ensures a label accuracy above 95% for every class and tight bounding boxes. Finally, we train and evaluate well-known deep network architectures and report baseline figures for per-frame classification and localization. We also demonstrate how the temporal contiguity of video can potentially be used to improve such inferences. The data set can be found at https://research.google.com/youtube-bb. We hope the availability of such large curated corpus will spur new advances in video object detection and tracking.

We present an interpretation of Inception modules in convolutional neural networks as being an intermediate step in-between regular convolution and the depthwise separable convolution operation (a depthwise convolution followed by a pointwise convolution). In this light, a depthwise separable convolution can be understood as an Inception module with a maximally large number of towers. This observation leads us to propose a novel deep convolutional neural network architecture inspired by Inception, where Inception modules have been replaced with depthwise separable convolutions. We show that this architecture, dubbed Xception, slightly outperforms Inception V3 on the ImageNet dataset (which Inception V3 was designed for), and significantly outperforms Inception V3 on a larger image classification dataset comprising 350 million images and 17,000 classes. Since the Xception architecture has the same number of parameters as Inception V3, the performance gains are not due to increased capacity but rather to a more efficient use of model parameters.

One of the most frequently applied low-level operations in computer vision is the conversion of an RGB camera image into its luminance representation. This is also one of the most incorrectly applied operations. Even our most trusted softwares, Matlab and OpenCV, do not perform luminance conversion correctly. In this paper, we examine the main factors that make proper RGB to luminance conversion difficult, in particular: 1) incorrect white-balance, 2) incorrect gamma/tone-curve correction, and 3) incorrect equations. Our analysis shows errors up to 50% for various colors are not uncommon. As a result, we argue that for most computer vision problems there is no need to attempt luminance conversion; instead, there are better alternatives depending on the task.

Underwater image reconstruction methods require the knowledge of wideband attenuation coefficients per color channel. Current estimation methods for these coefficients require specialized hardware or multiple images, and none of them leverage the multitude of existing ocean optical measurements as priors. Here, we aim to constrain the set of physically-feasible wideband attenuation coefficients in the ocean by utilizing water attenuation measured worldwide by oceanographers. We calculate the space of valid wideband effective attenuation coefficients in the 3D RGB domain and find that a bound manifold in 3-space sufficiently represents the variation from the clearest to murkiest waters. We validate our model using in situ experiments in two different optical water bodies, the Red Sea and the Mediterranean. Moreover, we show that contradictory to the common image formation model, the coefficients depend on the imaging range and object reflectance, and quantify the errors resulting from ignoring these dependencies.

Aggregating extra features has been considered as an effective approach to boost traditional pedestrian detection methods. However, there is still a lack of studies on whether and how CNN-based pedestrian detectors can benefit from these extra features. The first contribution of this paper is exploring this issue by aggregating extra features into CNN-based pedestrian detection framework. Through extensive experiments, we evaluate the effects of different kinds of extra features quantitatively. Moreover, we propose a novel network architecture, namely HyperLearner, to jointly learn pedestrian detection as well as the given extra feature. By multi-task training, HyperLearner is able to utilize the information of given features and improve detection performance without extra inputs in inference. The experimental results on multiple pedestrian benchmarks validate the effectiveness of the proposed HyperLearner.

Quantization is considered as one of the most effective methods to optimize the inference cost of neural network models for their deployment to mobile and embedded systems, which have tight resource constraints. In such approaches, it is critical to provide low-cost quantization under a tight accuracy loss constraint (e.g., 1%). In this paper, we propose a novel method for quantizing weights and activations based on the concept of weighted entropy. Unlike recent work on binary-weight neural networks, our approach is multi-bit quantization, in which weights and activations can be quantized by any number of bits depending on the target accuracy. This facilitates much more flexible exploitation of accuracy-performance trade-off provided by different levels of quantization. Moreover, our scheme provides an automated quantization flow based on conventional training algorithms, which greatly reduces the design-time effort to quantize the network. According to our extensive evaluations based on practical neural network models for image classification (AlexNet, GoogLeNet and ResNet-50/101), object detection (R-FCN with 50-layer ResNet), and language modeling (an LSTM network), our method achieves significant reductions in both the model size and the amount of computation with minimal accuracy loss. Also, compared to existing quantization schemes, ours provides higher accuracy with a similar resource constraint and requires much lower design effort.

We propose a weakly supervised semantic segmentation algorithm that uses image tags for supervision. We apply the tags in queries to collect three sets of web images, which encode the clean foregrounds, the common back- grounds, and realistic scenes of the classes. We introduce a novel three-stage training pipeline to progressively learn semantic segmentation models. We first train and refine a class-specific shallow neural network to obtain segmentation masks for each class. The shallow neural networks of all classes are then assembled into one deep convolutional neural network for end-to-end training and testing. Experiments show that our method notably outperforms previous state-of-the-art weakly supervised semantic segmentation approaches on the PASCAL VOC 2012 segmentation bench- mark. We further apply the class-specific shallow neural networks to object segmentation and obtain excellent results.

We propose a weakly-supervised approach that takes image-sentence pairs as input and learns to visually ground (i.e., localize) arbitrary linguistic phrases, in the form of spatial attention masks. Specifically, the model is trained with images and their associated image-level captions, without any explicit region-to-phrase correspondence annotations. To this end, we introduce an end-to-end model which learns visual groundings of phrases with two types of carefully designed loss functions. In addition to the standard discriminative loss, which enforces that attended image regions and phrases are consistently encoded, we propose a novel structural loss which makes use of the parse tree structures induced by the sentences. In particular, we ensure complementarity among the attention masks that correspond to sibling noun phrases, and compositionality of attention masks among the children and parent phrases, as defined by the sentence parse tree. We validate the effectiveness of our approach on the Microsoft COCO and Visual Genome datasets.

This paper focuses on a novel and challenging vision task, dense video captioning, which aims to automatically describe a video clip with multiple informative and diverse caption sentences. The proposed method is trained without explicit annotation of fine-grained sentence to video region-sequence correspondence, but is only based on weak video-level sentence annotations. It differs from existing video captioning systems in three technical aspects. First, we propose lexical fully convolutional neural networks (Lexical-FCN) with weakly supervised multi-instance multi-label learning to weakly link video regions with lexical labels. Second, we introduce a novel submodular maximization scheme to generate multiple informative and diverse region-sequences based on the Lexical-FCN outputs. A winner-takes-all scheme is adopted to weakly associate sentences to region-sequences in the training phase. Third, a sequence-to-sequence learning based language model is trained with the weakly supervised information obtained through the association process. We show that the proposed method can not only produce informative and diverse dense captions, but also outperform state-of-the-art single video captioning methods by a large margin.

Object detection is a challenging task in visual understanding domain, and even more so if the supervision is to be weak. Recently, few efforts to handle the task without expensive human annotations is established by promising deep neural network. A new architecture of cascaded networks is proposed to learn a convolutional neural network (CNN) under such conditions. We introduce two such architectures, with either two cascade stages or three which are trained in an end-to-end pipeline. The first stage of both architectures extracts best candidate of class specific region proposals by training a fully convolutional network. In the case of the three stage architecture, the middle stage provides object segmentation, using the output of the activation maps of first stage. The final stage of both architectures is a part of a convolutional neural network that performs multiple instance learning on proposals extracted in the previous stage(s). Our experiments on the PASCAL VOC 2007, 2010, 2012 and large scale object datasets, ILSVRC 2013, 2014 datasets show improvements in the areas of weakly-supervised object detection, classification and localization.

Localizing functional regions of objects or affordances is an important aspect of scene understanding and relevant for many robotics applications. In this work, we introduce a pixel-wise annotated affordance dataset of 3090 images containing 9916 object instances. Since parts of an object can have multiple affordances, we address this by a convo- lutional neural network for multilabel affordance segmen- tation. We also propose an approach to train the network from very few keypoint annotations. Our approach achieves a higher affordance detection accuracy than other weakly supervised methods that also rely on keypoint annotations or image annotations as weak supervision.

Fine-grained activity understanding in videos has attracted considerable recent attention with a shift from action classification to detailed actor and action understanding that provides compelling results for perceptual needs of cutting-edge autonomous systems. However, current methods for detailed understanding of actor and action have significant limitations: they require large amounts of finely labeled data, and they fail to capture any internal relationship among actors and actions. To address these issues, in this paper, we propose a novel, robust multi-task ranking model for weakly supervised actor-action segmentation where only video-level tags are given for training samples. Our model is able to share useful information among different actors and actions while learning a ranking matrix to select representative supervoxels for actors and actions respectively. Final segmentation results are generated by a conditional random field that considers various ranking scores for different video parts. Extensive experimental results on the Actor-Action Dataset (A2D) demonstrate that the proposed approach outperforms the state-of-the-art weakly supervised methods and performs as well as the top-performing fully supervised method.

Image-based precision medicine techniques can be used to better treat cancer patients. However, the gigapixel resolution of Whole Slide Histopathological Images (WSIs) makes traditional survival models computationally impossible. These models usually adopt manually labeled discriminative patches from region of interests (ROIs) and are unable to directly learn discriminative patches from WSIs. We argue that only a small set of patches cannot fully represent the patients' survival status due to the heterogeneity of tumor. Another challenge is that survival prediction usually comes with insufficient training patient samples. In this paper, we propose an effective Whole Slide Histopathological Images Survival Analysis framework (WSISA) to overcome above challenges. To exploit survival-discriminative patterns from WSIs, we first extract hundreds of patches from each WSI by adaptive sampling and then group these images into different clusters. Then we propose to train an aggregation model to make patient-level predictions based on cluster-level Deep Convolutional Survival (DeepConvSurv) prediction results. Different from existing state-of-the-arts image-based survival models which extract features using some patches from small regions of WSIs, the proposed framework can efficiently exploit and utilize all discriminative patterns in WSIs to predict patients' survival status. To the best of our knowledge, this has not been shown before. We apply our method to the survival predictions of glioma and non-small-cell lung cancer using three datasets. Results demonstrate the proposed framework can significantly improve the prediction performance compared with the existing state-of-the-arts survival methods.

This paper introduces WILDCAT, a deep learning method which jointly aims at aligning image regions for gaining spatial invariance and learning strongly localized features. Our model is trained using only global image labels and is devoted to three main visual recognition tasks: image classification, weakly supervised object localization and semantic segmentation. WILDCAT extends state-of-the-art Convolutional Neural Networks at three main levels: the use of Fully Convolutional Networks for maintaining spatial resolution, the explicit design in the network of local features related to different class modalities, and a new way to pool these features to provide a global image prediction required for weakly supervised training. Extensive experiments show that our model significantly outperforms state-of-the-art methods.

We describe a system to detect objects in three-dimensional space using video and inertial sensors (accelerometer and gyrometer), ubiquitous in modern mobile platforms from phones to drones. Inertials afford the ability to impose class-specific scale priors for objects, and provide a global orientation reference. A minimal sufficient representation, the posterior of semantic (identity) and syntactic (pose) attributes of objects in space, can be decomposed into a geometric term, which can be maintained by a localization-and-mapping filter, and a likelihood function, which can be approximated by a discriminatively-trained convolutional neural network The resulting system can process the video stream causally in real time, and provides a representation of objects in the scene that is persistent: Confidence in the presence of objects grows with evidence, and objects previously seen are kept in memory even when temporarily occluded, with their return into view automatically predicted to prime re-detection.