We propose novel Stacked Spatio-Temporal Graph Convolutional Networks (Stacked-STGCN) for action segmentation, i.e., predicting and localizing a sequence of actions over long videos. We extend the Spatio-Temporal Graph Convolutional Network (STGCN) originally proposed for skeleton-based action recognition to enable nodes with different characteristics (e.g., scene, actor, object, action, etc.), feature descriptors with varied lengths, and arbitrary temporal edge connections to account for large graph deformation commonly associated with complex activities. We further introduce the stacked hourglass architecture to STGCN to leverage the advantages of an encoder-decoder design for improved generalization performance and localization accuracy. We explore various descriptors such as frame-level VGG, segment-level I3D, RCNN-based object, etc. as node descriptors to enable action segmentation based on joint inference over comprehensive contextual information. We show results on CAD120 (which provides pre-computed node features and edge weights for fair performance comparison across algorithms) as well as a more complex real-world activity dataset, Charades. Our Stacked-STGCN in general achieves 4.0% performance improvement over the best reported results in F1 score on CAD120 and 1.3% in mAP on Charades using VGG features.
This paper describes AutoFocus, an efficient multi-scale inference algorithm for deep-learning based object detectors. Instead of processing an entire image pyramid, AutoFocus adopts a coarse to fine approach and only processes regions which are likely to contain small objects at finer scales. This is achieved by predicting category agnostic segmentation maps for small objects at coarser scales, called FocusPixels. FocusPixels can be predicted with high recall, and in many cases, they only cover a small fraction of the entire image. To make efficient use of FocusPixels, an algorithm is proposed which generates compact rectangular FocusChips which enclose FocusPixels. The detector is only applied inside FocusChips, which reduces computation while processing finer scales. Different types of error can arise when detections from FocusChips of multiple scales are combined, hence techniques to correct them are proposed. AutoFocus obtains an mAP of 47.9% (68.3% at 50% overlap) on the COCO test-dev set while processing 6.4 images per second on a Titan X (Pascal) GPU. This is 2.5X faster than our multi-scale baseline detector and matches its mAP. The number of pixels processed in the pyramid can be reduced by 5X with a 1% drop in mAP. AutoFocus obtains more than 10% mAP gain compared to RetinaNet but runs at the same speed with the same ResNet-101 backbone.
This research strives for natural language moment retrieval in long, untrimmed video streams. The problem nevertheless is not trivial especially when a video contains multiple moments of interests and the language describes complex temporal dependencies, which often happens in real scenarios. We identify two crucial challenges: semantic misalignment and structural misalignment. However, existing approaches treat different moments separately and do not explicitly model complex moment-wise temporal relations. In this paper, we present Moment Alignment Network (MAN), a novel framework that unifies the candidate moment encoding and temporal structural reasoning in a single-shot feed-forward network. MAN naturally assigns candidate moment representations aligned with language semantics over different temporal locations and scales. Most importantly, we propose to explicitly model moment-wise temporal relations as a structured graph and devise an iterative graph adjustment network to jointly learn the best structure in an end-to-end manner. We evaluate the proposed approach on two challenging public benchmarks Charades-STA and DiDeMo, where our MAN significantly outperforms the state-of-the-art by a large margin.
We present AdaFrame, a framework that adaptively selects relevant frames on a per-input basis for fast video recognition. AdaFrame contains a Long Short-Term Memory network augmented with a global memory that provides context information for searching which frames to use over time. Trained with policy gradient methods, AdaFrame generates a prediction, determines which frame to observe next, and computes the utility, i.e., expected future rewards, of seeing more frames at each time step. At testing time, AdaFrame exploits predicted utilities to achieve adaptive lookahead inference such that the overall computational costs are reduced without incurring a decrease in accuracy. Extensive experiments are conducted on two large-scale video benchmarks, FCVID and AvtivityNet. AdaFrame matches the performance of using all frames with only 8.21 and 8.65 frames on FCVID and AvtivityNet, respectively. We further qualitatively demonstrate learned frame usage can indicate the difficulty of making classification decisions; easier samples need fewer frames while harder ones require more, both at instance-level within the same class and at class-level among different categories.
Standard adversarial attacks change the predicted class label of an image by adding specially tailored small perturbations to its pixels. In contrast, a universal perturbation is an update that can be added to any image in a broad class of images, while still changing the predicted class label. We study the efficient generation of universal adversarial perturbations, and also efficient methods for hardening networks to these attacks. We propose a simple optimization-based universal attack that reduces the top-1 accuracy of various network architectures on ImageNet to less than 20%, while learning the universal perturbation 13X faster than the standard method. To defend against these perturbations, we propose universal adversarial training, which models the problem of robust classifier generation as a two-player min-max game. This method is much faster and more scalable than conventional adversarial training with a strong adversary (PGD), and yet yields models that are extremely resistant to universal attacks, and comparably resistant to standard (per-instance) black box attacks. We also discover a rather fascinating side-effect of universal adversarial training: attacks built for universally robust models transfer better to other (black box) models than those built with conventional adversarial training.
It has been witnessed an emerging demand for image manipulation segmentation to distinguish between fake images produced by advanced photo editing software and authentic ones. In this paper, we describe an approach based on semantic segmentation for detecting image manipulation. The approach consists of three stages. A generation stage generates hard manipulated images from authentic images using a Generative Adversarial Network (GAN) based model by cutting a region out of a training sample, pasting it into an authentic image and then passing the image through a GAN to generate harder true positive tampered region. A segmentation stage and a replacement stage, sharing weights with each other, then collaboratively construct dense predictions of tampered regions. We achieve state-of-the-art performance on four public image manipulation detection benchmarks while maintaining robustness to various attacks.
Researchers have observed that Visual Question Answering (VQA) models tend to answer questions by learning statistical biases in the data. For example, their answer to the question "What is the color of the grass?" is usually "Green", whereas a question like "What is the title of the book?" cannot be answered by inferring statistical biases. It is of interest to the community to explicitly discover such biases, both for understanding the behavior of such models, and towards debugging them. Our work address this problem. In a database, we store the words of the question, answer and visual words corresponding to regions of interest in attention maps. By running simple rule mining algorithms on this database, we discover human-interpretable rules which give us unique insight into the behavior of such models. Our results also show examples of unusual behaviors learned by models in attempting VQA tasks.
Recent advances in deep convolutional neural networks (CNNs) have motivated researchers to adapt CNNs to directly model points in 3D point clouds. Modeling local structure has been proven to be important for the success of convolutional architectures, and researchers exploited the modeling of local point sets in the feature extraction hierarchy. However, limited attention has been paid to explicitly model the geometric structure amongst points in a local region. To address this problem, we propose Geo-CNN, which applies a generic convolution-like operation dubbed as GeoConv to each point and its local neighborhood. Local geometric relationships among points are captured when extracting edge features between the center and its neighboring points. We first decompose the edge feature extraction process onto three orthogonal bases, and then aggregate the extracted features based on the angles between the edge vector and the bases. This encourages the network to preserve the geometric structure in Euclidean space throughout the feature extraction hierarchy. GeoConv is a generic and efficient operation that can be easily integrated into 3D point cloud analysis pipelines for multiple applications. We evaluate Geo-CNN on ModelNet40 and KITTI and achieve state-of-the-art performance.
Most work on temporal action detection is formulated in an offline manner, in which the start and end times of actions are determined after the entire video is fully observed. However, real-time applications including surveillance and driver assistance systems require identifying actions as soon as each video frame arrives, based only on current and historical observations. In this paper, we propose a novel framework, Temporal Recurrent Networks (TRNs), to model greater temporal context of a video frame by simultaneously performing online action detection and anticipation of the immediate future. At each moment in time, our approach makes use of both accumulated historical evidence and predicted future information to better recognize the action that is currently occurring, and integrates both of these into a unified end-to-end architecture. We evaluate our approach on two popular online action detection datasets, HDD and TVSeries, as well as another widely used dataset, THUMOS'14. The results show that TRN significantly outperforms the state-of-the-art.
Existing 3D pose datasets of object categories are limited to generic object types and lack of fine-grained information. In this work, we introduce a new large-scale dataset that consists of 409 fine-grained categories and 31,881 images with accurate 3D pose annotation. Specifically, we augment three existing fine-grained object recognition datasets (StanfordCars, CompCars and FGVC-Aircraft) by finding a specific 3D model for each sub-category from ShapeNet and manually annotating each 2D image by adjusting a full set of 7 continuous perspective parameters. Since the fine-grained shapes allow 3D models to better fit the images, we further improve the annotation quality by initializing from the human annotation and conducting local search of the pose parameters with the objective of maximizing the IoUs between the projected mask and the segmentation reference estimated from state-of-the-art deep Convolutional Neural Networks (CNNs). We provide full statistics of the annotations with qualitative and quantitative comparisons suggesting that our dataset can be a complementary source for studying 3D pose estimation. The dataset can be downloaded at http://users.umiacs.umd.edu/~wym/3dpose.html.