Motion plays a crucial role in understanding videos and most state-of-the-art neural models for video classification incorporate motion information typically using optical flows extracted by a separate off-the-shelf method. As the frame-by-frame optical flows require heavy computation, incorporating motion information has remained a major computational bottleneck for video understanding. In this work, we replace external and heavy computation of optical flows with internal and light-weight learning of motion features. We propose a trainable neural module, dubbed MotionSqueeze, for effective motion feature extraction. Inserted in the middle of any neural network, it learns to establish correspondences across frames and convert them into motion features, which are readily fed to the next downstream layer for better prediction. We demonstrate that the proposed method provides a significant gain on four standard benchmarks for action recognition with only a small amount of additional cost, outperforming the state of the art on Something-Something-V1&V2 datasets.
论文检索
输入标题、作者或关键词，从 12,319 篇学术成果中精准定位
A key capability of an intelligent system is deciding when events from past experience must be remembered and when they can be forgotten. Towards this goal, we develop a predictive model of human visual event memory and how those memories decay over time. We introduce Memento10k, a new, dynamic video memorability dataset containing human annotations at different viewing delays. Based on our findings, we propose a new mathematical formulation of memorability decay, resulting in a model that is able to produce the first quantitative estimation of how a video decays in memory over time. In contrast with previous work, our model can predict the probability that a video will be remembered at an arbitrary delay. Importantly, our approach combines visual and semantic information (in the form of textual captions) to fully represent the meaning of events. Our experiments on two video memorability benchmarks, including Memento10k, show that our model significantly improves upon the best prior approach (by 12% on average).
We present a novel method to reconstruct the 3D layout of a room—walls, floors, ceilings—from a single perspective view in challenging conditions, by contrast with previous single-view methods restricted to cuboid-shaped layouts. This input view can consist of a color image only, but considering a depth map results in a more accurate reconstruction. Our approach is formalized as solving a constrained discrete optimization problem to find the set of 3D polygons that constitute the layout. In order to deal with occlusions between components of the layout, which is a problem ignored by previous works, we introduce an analysis-by-synthesis method to iteratively refine the 3D layout estimate. As no dataset was available to evaluate our method quantitatively, we created one together with several appropriate metrics. Our dataset consists of 293 images from ScanNet, which we annotated with precise 3D layouts. It offers three times more samples than the popular NYUv2 303 benchmark, and a much larger variety of layouts.
Requirement of large annotated datasets restricts the use of deep convolutional neural networks (CNNs) for many practical applications. The problem can be mitigated by using active learning (AL) techniques which, under a given annotation budget, allow to select a subset of data that yields maximum accuracy upon fine tuning. State of the art AL approaches typically relies on measures of visual diversity or prediction uncertainty, which are unable to effectively capture the variations in the spatial context. On the other hand, modern CNN architectures make heavy use of spatial context for achieving highly accurate predictions. Since the context is difficult to evaluate in the absence of ground-truth labels, we introduce the notion of contextual diversity that captures the confusion associated with spatially co-occurring classes. Contextual Diversity (CD) hinges on a crucial observation that the probability vector predicted by a CNN for a region of interest typically contains information from a larger receptive field. Exploiting this observation, we use the proposed CD measure within two AL frameworks: (1) a core-set based strategy and (2) a reinforcement learning based policy, for the active frame selection. Our extensive empirical evaluation establishes state of the art results for active learning on benchmark datasets of Semantic Segmentation, Object Detection and Image classification. Our ablation studies show clear advantages of using contextual diversity for active learning.
Modern methods for counting people in crowded scenes rely on deep networks to estimate people densities in individual images. As such, only very few take advantage of temporal consistency in video sequences, and those that do only impose weak smoothness constraints across consecutive frames. In this paper, we advocate estimating people flows across image locations between consecutive images and inferring the people densities from these flows instead of directly regressing. This enables us to impose much stronger constraints encoding the conservation of the number of people. As a result, it significantly boosts performance without requiring a more complex architecture. Furthermore, it also enables us to exploit the correlation between people flow and optical flow to further improve the results. We will demonstrate that we consistently outperform state-of-the-art methods on five benchmark datasets.
We address the curve lane detection problem which poses more real-world challenges than conventional lane detection for better facilitating modern assisted/autonomous driving systems. Current hand-designed lane detection methods are not robust enough to capture the curve lanes especially the remote ones due to the lack of modeling both long-range contextual information and detailed curve trajectory. In this paper, we propose a novel lane-sensitive architecture search framework named CurveLane-NAS to automatically capture both long-ranged coherent and accurate short-range curve information while unifying both architecture search and post-processing on curve lane predictions via point blending. It consists of three search modules: a) a feature fusion search module to find a better fusion of the local and global context for multi-level hierarchy features; b) an elastic backbone search module to explore an efficient feature extractor with good semantics and latency; c) an adaptive point blending module to search a multi-level post-processing refinement strategy to combine multi-level head prediction. The unified framework ensures lane-sensitive predictions by the mutual guidance between NAS and adaptive point blending. Furthermore, we also steer forward to release a more challenging benchmark named CurveLanes for addressing the most difficult curve lanes. It consists of 150K images with 680K labels. Experiments on the new CurveLanes show that the SOTA lane detection methods suffer substantial performance drop while our model can still reach an 80+% F1-score. Extensive experiments on traditional lane benchmarks such as CULane also demonstrate the superiority of our CurveLane-NAS, e.g. achieving a new SOTA result 74.8% of F1-score on CULane dataset.
We introduce OmniSource, a novel framework for leveraging web data to train video recognition models. OmniSource overcomes the barriers between data formats, such as images, short videos, and long untrimmed videos for webly-supervised learning. First, data samples with multiple formats, curated by task-specific data collection and automatically filtered by a teacher model, are transformed into a unified form. Then a joint-training strategy is proposed to deal with the domain gaps between multiple data sources and formats in webly-supervised learning. Several good practices, including data balancing, resampling, and cross-dataset mixup are adopted in joint training. Experiments show that by utilizing data from multiple sources and formats, OmniSource is more data-efficient in training. With only 3.5M images and 800K minutes videos crawled from the internet without human labeling (less than 2% of prior works), our models learned with OmniSource improve Top-1 accuracy of 2D- and 3D-ConvNet baseline models by 3.0% and 3.9%, respectively, on the Kinetics-400 benchmark. With OmniSource, we establish new records with different pretraining strategies for video recognition. Our best models achieve 80.4%, 80.5%, and 83.6% Top-1 accuracies on the Kinetics-400 benchmark respectively for training-from-scratch, ImageNet pre-training and IG-65M pre-training.
Weakly supervised object localization (WSOL) is a task of localizing an object in an image only using image-level labels. To tackle the WSOL problem, most previous studies have followed the conventional class activation mapping (CAM) pipeline: (i) training CNNs for a classification objective, (ii) generating a class activation map via global average pooling (GAP) on feature maps, and (iii) extracting bounding boxes by thresholding based on the maximum value of the class activation map. In this work, we reveal the current CAM approach suffers from three fundamental issues: (i) the bias of GAP that assigns a higher weight to a channel with a small activation area, (ii) negatively weighted activations inside the object regions and (iii) instability from the use of the maximum value of a class activation map as a thresholding reference. They collectively cause the problem that the localization to be highly limited to small regions of an object. We propose three simple but robust techniques that alleviate the problems, including thresholded average pooling, negative weight clamping, and percentile as a standard for thresholding. Our solutions are universally applicable to any WSOL methods using CAM and improve their performance drastically. As a result, we achieve the new state-of-the-art performance on three benchmark datasets of CUB-200-2011, ImageNet-1K, and OpenImages30K.
We propose a unified referring video object segmentation network (URVOS). URVOS takes a video and a referring expression as inputs, and estimates the {object masks} referred by the given language expression in the whole video frames. Our algorithm addresses the challenging problem by performing language-based object segmentation and mask propagation jointly using a single deep neural network with a proper combination of two attention models. In addition, we construct the first large-scale referring video object segmentation dataset called Refer-Youtube-VOS. We evaluate our model on two benchmark datasets including ours and demonstrate the effectiveness of the proposed approach. The dataset is released at \url{https://github.com/skynbe/Refer-Youtube-VOS}.
Most existing object instance detection and segmentation models only work well on fairly balanced benchmarks where per-category training sample numbers are comparable, such as COCO. They tend to suffer performance drop on realistic datasets that are usually long-tailed. This work aims to study and address such open challenges. Specifically, we systematically investigate performance drop of the state-of-the-art two-stage instance segmentation model Mask R-CNN on the recent long-tail LVIS dataset, and unveil that a major cause is the inaccurate classification of object proposals. Based on such an observation, we first consider various techniques for improving long-tail classification performance which indeed enhance instance segmentation results. We then propose a simple calibration framework to more effectively alleviate classification head bias with a bi-level class balanced sampling approach. Without bells and whistles, it significantly boosts the performance of instance segmentation for tail classes on the recent LVIS dataset and our sampled COCO-LT dataset. Our analysis provides useful insights for solving long-tail instance detection and segmentation problems, and the straightforward mph{SimCal} method can serve as a simple but strong baseline. With the method we have won the 2019 LVIS challenge. Codes and models are available at \url{https://github.com/twangnh/SimCal}.
Knowledge Distillation (KD) has been one of the most popular used methods to learn a compact model. However, it still suffers from high demand in time and computational resources caused by sequential training pipeline. Furthermore, the soft targets from deeper models do not often serve as good cues for the shallower models due to the gap of compatibility. In this work, we consider these two problems at the sametime. Specifically, we propose that better soft targets with higher compatibility can be generated by using a label generator to fuse the featuremaps from deeper stages in a top-down manner, and we can employ the meta-learning technique to optimize this label generator. Utilizing the soft targets learned from the intermediate feature maps of the model, we can achieve better self-boosting of the network in comparison with the state-of-the-art. The experiments are conducted on two standard classi-fication benchmarks, namely CIFAR-100 and ILSVRC2012. We test various network architectures to show the generalizability of our MetaDistiller. The experiments results on two datasets strongly demonstrate theeffectiveness of our method.
Attract, Perturb, and Explore: Learning a Feature Alignment Network for Semi-supervised Domain Adaptation
PDF ↗Perturb, and Explore: Learning a Feature Alignment Network for Semi-supervised Domain Adaptation","Although unsupervised domain adaptation methods have been widely adopted across several computer vision tasks, it is more desirable if we can exploit a few labeled data from new domains encountered in a real application. The novel setting of the semi-supervised domain adaptation (SSDA) problem shares the challenges with the domain adaptation problem and the semi-supervised learning problem. However, a recent study shows that conventional domain adaptation and semi-supervised learning methods often result in less effective or negative transfer in the SSDA problem. In order to interpret the observation and address the SSDA problem, in this paper, we raise the intra-domain discrepancy issue within the target domain, which has never been discussed so far. Then, we demonstrate that addressing the intra-domain discrepancy leads to the ultimate goal of the SSDA problem. We propose an SSDA framework that aims to align features via alleviation of the intra-domain discrepancy. Our framework mainly consists of three schemes, i.e., attraction, perturbation, and exploration. First, the attraction scheme globally minimizes the intra-domain discrepancy within the target domain. Second, we demonstrate the incompatibility of the conventional adversarial perturbation methods with SSDA. Then, we present a domain adaptive adversarial perturbation scheme, which perturbs the given target samples in a way that reduces the intra-domain discrepancy. Finally, the exploration scheme locally aligns features in a class-wise manner complementary to the attraction scheme by selectively aligning unlabeled target features complementary to the perturbation scheme. We conduct extensive experiments on domain adaptation benchmark datasets such as DomainNet, Office-Home, and Office. Our method achieves state-of-the-art performances on all datasets.
Object tracking is a well-studied problem in computer vision while identifying salient spots of objects in a video is a less explored direction in the literature. Video eye gaze estimation methods aim to tackle a related task but salient spots in those methods are not bounded by objects and tend to produce very scattered, unstable predictions due to the noisy ground truth data. We reformulate the problem of detecting and tracking of salient object spots as a new task called object hotspot tracking. In this paper, we propose to tackle this task jointly with unsupervised video object segmentation, in real-time, with a unified framework to exploit the synergy between the two. Specifically, we propose a Weighted Correlation Siamese Network (WCS-Net) which employs a Weighted Correlation Block (WCB) for encoding the pixel-wise correspondence between a template frame and the search frame. In addition, WCB takes the initial mask / hotspot as guidance to enhance the influence of salient regions for robust tracking. Our system can operate online during inference and jointly produce the object mask and hotspot track-lets at 33 FPS. Experimental results validate the effectiveness of our network design, and show the benefits of jointly solving the hotspot tracking and object segmentation problems. In particular, our method performs favorably against state-of-the-art video eye gaze models in object hotspot tracking, and outperforms existing methods on three benchmark datasets for unsupervised video object segmentation.
Video interpolation increases the temporal resolution of a video sequence by synthesizing intermediate frames between two consecutive frames. We propose a novel deep-learning-based video interpolation algorithm based on bilateral motion estimation. First, we develop the bilateral motion network with the bilateral cost volume to estimate bilateral motions accurately. Then, we approximate bi-directional motions to predict a different kind of bilateral motions. We then warp the two input frames using the estimated bilateral motions. Next, we develop the dynamic filter generation network to yield dynamic blending filters. Finally, we combine the warped frames using the dynamic blending filters to generate intermediate frames. Experimental results show that the proposed algorithm outperforms the state-of-the-art video interpolation algorithms on several benchmark datasets.
Photo retouching aims at enhancing the aesthetic visual quality of images that suffer from photographic defects such as over/under exposure, poor contrast, inharmonious saturation. Practically, photo retouching can be accomplished by a series of image processing operations. In this paper, we investigate some commonly-used retouching operations and mathematically find that these pixel-independent operations can be approximated or formulated by multi-layer perceptrons (MLPs). Based on this analysis, we propose an extremely light-weight framework - Conditional Sequential Retouching Network (CSRNet) - for efficient global image retouching. CSRNet consists of a base network and a condition network. The base network acts like an MLP that processes each pixel independently and the condition network extracts the global features of the input image to generate a condition vector. To realize retouching operations, we modulate the intermediate features using Global Feature Modulation (GFM), of which the parameters are transformed by condition vector. Benefiting from the utilization of $1 imes1$ convolution, CSRNet only contains less than 37k trainable parameters, which is orders of magnitude smaller than existing learning-based methods. Extensive experiments show that our method achieves state-of-the-art performance on the benchmark MIT-Adobe FiveK dataset quantitively and qualitatively. Code is available at \url{https://github.com/hejingwenhejingwen/CSRNet}.
Continual learning from a sequential stream of data is a crucial challenge for machine learning research. Most studies have been conducted on this topic under the single-label classification setting along with an assumption of balanced label distribution. This work expands this research horizon towards multi-label classification. In doing so, we identify unanticipated adversity innately existent in many multilabel datasets, the long-tailed distribution. We jointly address the two independently solved problems, Catastropic Forgetting and the long-tailed label distribution by first empirically showing a new challenge of destructive forgetting of the minority concepts on the tail. Then, we curate two benchmark datasets, COCOseq and NUS-WIDEseq, that allow the study of both intra- and inter-task imbalances. Lastly, we propose a new sampling strategy for replay-based approach named Partitioning Reservoir Sampling (PRS), which allows the model to maintain a balanced knowledge of both head and tail classes. We publicly release the dataset and the code in our project page.
The recently proposed pseudo-LiDAR based 3D detectors greatly improves the benchmark of monocular/stereo 3D detection task. However, the underlying mechanism is still obscure to the research community. In this paper, we perform an in-depth investigation and observe that the pseudo-LiDAR representation is effective because of the coordinate transformation, instead of data representation itself. Based on this observation, we design an image based CNN detector named PatchNet, which is a generalized version that can represent pseudo-LiDAR based 3D detectors. In PatchNet, we organize pseudo-LiDAR data as the image representation, which means existing 2D CNN designs can be easily utilized for extracting deep features from input data and boosting 3D detection performance. We conduct extensive experiments on the challenging KITTI dataset, where the proposed PatchNet outperforms all existing pseudo-LiDAR based counterparts. Co de has been made available at: https://github.com/xinzhuma/patchnet
This paper studies the problem of predicting future trajectories of people in unseen cameras of novel scenarios and views. We approach this problem through the real-data-free setting in which the model is trained only on 3D simulation data and applied out-of-the-box to a wide variety of real cameras. We propose a novel approach to learn robust representation through augmenting the simulation training data such that the representation can better generalize to unseen real-world test data. The key idea is to mix the feature of the hardest camera view with the adversarial feature of the original view. We refer to our method as $ extit{SimAug}$. We show that $ extit{SimAug}$ achieves promising results on three real-world benchmarks using zero real training data, and state-of-the-art performance in the Stanford Drone and the VIRAT/ActEV dataset when using in-domain training data. Code and models are released at https://next.cs.cmu.edu/simaug
Recent studies have shown that current VQA models are heavily biased on the language priors in the train set to answer the question, irrespective of the image. E.g., overwhelmingly answer ""what sport is"" as ""tennis"" or ""what color banana"" as ""yellow."" This behavior restricts them from real-world application scenarios. In this work, we propose a novel model-agnostic question encoder, Visually-Grounded Question Encoder (VGQE), for VQA that reduces this effect. VGQE utilizes both visual and language modalities equally while encoding the question. Hence the question representation itself gets sufficient visual-grounding, and thus reduces the dependency of the model on the language priors. We demonstrate the effect of VGQE on three recent VQA models and achieve state-of-the-art results on the bias-sensitive split of the VQAv2 dataset; VQA-CPv2. Further, unlike the existing bias-reduction techniques, on the standard VQAv2 benchmark, our approach does not drop the accuracy; instead, it improves the performance.
Estimating eye-gaze from images alone is a challenging task, in large parts due to un-observable person-specific factors. Achieving high accuracy typically requires labeled data from test users which may not be attainable in real applications. We observe that there exists a strong relationship between what users are looking at and the appearance of the user’s eyes. In response to this understanding, we propose a novel dataset and accompanying method which aims to explicitly learn these semantic and temporal relationships. Our video dataset consists of time-synchronized screen recordings, user-facing camera views, and eye gaze data, which allows for new benchmarks in temporal gaze tracking as well as label-free refinement of gaze. Importantly, we demonstrate that the fusion of information from visual stimuli as well as eye images can lead towards achieving performance similar to literature-reported figures acquired through supervised personalization. Our final method yields significant performance improvements on our proposed EVE dataset, with up to 28% improvement in Point-of-Gaze estimates (resulting in 2.49-deg in angular error), paving the path towards high-accuracy screen-based eye tracking purely from webcam sensors. The dataset and reference source code are available at https://ait.ethz.ch/projects/2020/EVE