Most recent approaches for online action detection tend to apply Recurrent Neural Network (RNN) to capture long-range temporal structure. However, RNN suffers from non-parallelism and gradient vanishing, hence it is hard to be optimized. In this paper, we propose a new encoder-decoder framework based on Transformers, named OadTR, to tackle these problems. The encoder attached with a task token aims to capture the relationships and global interactions between historical observations. The decoder extracts auxiliary information by aggregating anticipated future clip representations. Therefore, OadTR can recognize current actions by encoding historical information and predicting future context simultaneously. We extensively evaluate the proposed OadTR on three challenging datasets: HDD, TVSeries, and THUMOS14. The experimental results show that OadTR achieves higher training and inference speeds than current RNN based approaches, and significantly outperforms the state-of-the-art methods in terms of both mAP and mcAP. Code is available at https: //github.com/wangxiang1230/OadTR.
论文检索
输入标题、作者或关键词，从 2,893 篇学术成果中精准定位
We introduce a method to render Neural Radiance Fields (NeRFs) in real time using PlenOctrees, an octree-based 3D representation which supports view-dependent effects. Our method can render 800x800 images at more than 150 FPS, which is over 3000 times faster than conventional NeRFs. We do so without sacrificing quality while preserving the ability of NeRFs to perform free-viewpoint rendering of scenes with arbitrary geometry and view-dependent effects. Real-time performance is achieved by pre-tabulating the NeRF into a PlenOctree. In order to preserve view-dependent effects such as specularities, we factorize the appearance via closed-form spherical basis functions. Specifically, we show that it is possible to train NeRFs to predict a spherical harmonic representation of radiance, removing the viewing direction as an input to the neural network. Furthermore, we show that PlenOctrees can be directly optimized to further minimize the reconstruction loss, which leads to equal or better quality compared to competing methods. Moreover, this octree optimization step can be used to reduce the training time, as we no longer need to wait for the NeRF training to converge fully. Our real-time neural rendering approach may potentially enable new applications such as 6-DOF industrial and product visualizations, as well as next generation AR/VR systems. PlenOctrees are amenable to in-browser rendering as well; please visit the project page for the interactive online demo, as well as video and code: https://alexyu.net/plenoctrees.
Most state-of-the-art deep neural networks are overparameterized and exhibit a high computational cost. A straightforward approach to this problem is to replace convolutional kernels with its low-rank tensor approximations, whereas the Canonical Polyadic tensor Decomposition is one of the most suited models. However, fitting the convolutional tensors by numerical optimization algorithms often encounters diverging components, i.e., extremely large rank-one tensors but canceling each other. Such degeneracy often causes the non-interpretable result and numerical instability for the neural network fine-tuning. This paper is the first study on degeneracy in the tensor decomposition of convolutional kernels. We present a novel method, which can stabilize the low-rank approximation of convolutional kernels and ensure efficient compression while preserving the high-quality performance of the neural networks. We evaluate our approach on popular CNN architectures for image classification and show that our method results in much lower accuracy degradation and provides consistent performance.
We present a differentiable joint pruning and quantization (DJPQ) scheme. We frame neural network compression as a joint gradient-based optimization problem, trading off between model pruning and quantization automatically for hardware efficiency. DJPQ incorporates variational information bottleneck based structured pruning and mixed-bit precision quantization into a single differentiable loss function. While previous works always consider pruning and quantization separately, our method enables users to find the optimal trade-off between both in a single training procedure. To utilize the method for more efficient hardware inference, we extend DJPQ to integrate structured pruning with power-of-two bit-restricted quantization. We show that DJPQ significantly reduces the number of Bit-Operations (BOPs) for several networks while maintaining the top-1 accuracy of original floating-point models (e.g.,53x BOPs reduction in ResNet18 on ImageNet, 43x in MobileNetV2). Compared to the conventional two-stage approach, which optimizes pruning and quantization independently, our scheme outperforms in terms of both accuracy and BOPs. Even when considering bit-restricted quantization, DJPQ achieves larger compression ratios and better accuracy than the two-stage approach.
Curriculum learning can improve neural network training by guiding the optimization to desirable optima. We propose a novel curriculum learning approach for image classification that adapts the loss function by changing the label representation. The idea is to use a probability distribution over classes as target label, where the class probabilities reflect the similarity to the true class. Gradually, this label representation is shifted towards the standard one-hot-encoding. That is, in the beginning minor mistakes are corrected less than large mistakes, resembling a teaching process in which broad concepts are explained first before subtle differences are taught. The class similarity can be based on prior knowledge. For the special case of the labels being natural words, we propose a generic way to automatically compute the similarities. The natural words are embedded into Euclidean space using a standard word embedding. The probability of each class is then a function of the cosine similarity between the vector representations of the class and the true label. The proposed label-similarity curriculum learning (LCL) approach was empirically evaluated using several popular deep learning architectures for image classification tasks applied to five datasets including ImageNet, CIFAR100, and AWA2. In all scenarios, LCL was able to improve the classification accuracy on the test data compared to standard training. Code to reproduce results is available at https://github.com/speedystream/LCL.
Scene text detection has been significantly advanced over recent years, especially after the emergence of deep neural network. However, due to high diversity of scene texts in scale, orientation, shape and aspect ratio, as well as the inherent limitation of convolutional neural network for geometric transformations, to achieve accurate scene text detection is still an open problem. In this paper, we propose a novel sequential deformation method to effectively model the line-shape of scene text. An auxiliary character counting supervision is further introduced to guide the sequential offset prediction. The whole network can be easily optimized through an end-to-end multi-task manner. Extensive experiments are conducted on public scene text detection datasets including ICDAR 2017 MLT, ICDAR 2015, Total-text and SCUT-CTW1500. The experimental results demonstrate that the proposed method has outperformed previous state-of-the-art methods.
Self-driving cars need to understand 3D scenes efficiently and accurately in order to drive safely. Given the limited hardware resources, existing 3D perception models are not able to recognize small instances (e.g., pedestrians, cyclists) very well due to the low-resolution voxelization and aggressive downsampling. To this end, we propose Sparse Point-Voxel Convolution (SPVConv), a lightweight 3D module that equips the vanilla Sparse Convolution with the high-resolution point-based branch. With negligible overhead, this point-based branch is able to preserve the fine details even from large outdoor scenes. To explore the spectrum of efficient 3D models, we first define a flexible architecture design space based on SPVConv, and we then present 3D Neural Architecture Search (3D-NAS) to search the optimal network architecture over this diverse design space efficiently and effectively. Experimental results validate that the resulting SPVNAS model is fast and accurate: it outperforms the state-of-the-art MinkowskiNet by 3.3%, ranking 1st on the competitive SemanticKITTI leaderboard. It also achieves 8-24× computation reduction and 3-35× measured speedup over MinkowskiNet and KPConv with higher accuracy. Finally, we transfer our method to 3D object detection, and it achieves consistent improvements over the one-stage detection baseline on KITTI.
We propose and study a method called FLOT that estimates scene flow on point clouds. We start the design of FLOT by noticing that scene flow estimation on point clouds reduces to estimating a permutation matrix in a perfect world. Inspired by recent works on graph matching, we build a method to find these correspondences by borrowing tools from optimal transport. Then, we relax the transport constraints to take into account real-world imperfections. The transport cost between two points is given by the pairwise similarity between deep features extracted by a neural network trained under full supervision using synthetic datasets. Our main finding is that FLOT can perform as well as the best existing methods on synthetic and real-world datasets while requiring much less parameters and without using multiscale analysis. Our second finding is that, on the training datasets considered, most of the performance can be explained by the learned transport cost. This yields a simpler method, FLOT$_0$, which is obtained using a particular choice of optimal transport parameters and performs nearly as well as FLOT.
We present a visual localization framework based on novel deep attention aware features for autonomous driving that achieves centimeter level localization accuracy. Conventional approaches to the visual localization problem rely on handcrafted features or human-made objects on the road. They are known to be either prone to unstable matching caused by severe appearance or lighting changes, or too scarce to deliver constant and robust localization results in challenging scenarios. In this work, we seek to exploit the deep attention mechanism to search for salient, distinctive and stable features that are good for long-term matching in the scene through a novel end-to-end deep neural network. Furthermore, our learned feature descriptors are demonstrated to be competent to establish robust matches and therefore successfully estimate the optimal camera poses with high precision. We comprehensively validate the effectiveness of our method using a freshly collected dataset with high-quality ground truth trajectories and hardware synchronization between sensors. Results demonstrate that our method achieves a competitive localization accuracy when compared to the LiDAR-based localization solutions under various challenging circumstances, leading to a potential low-cost localization solution for autonomous driving.
Efficient search is a core issue in Neural Architecture Search (NAS). It is difficult for conventional NAS algorithms to directly search the architectures on large-scale tasks like ImageNet. In general, the cost of GPU hours for NAS grows with regard to training dataset size and candidate set size. One common way is searching on a smaller proxy dataset (e.g., CIFAR-10) and then transferring to the target task (e.g., ImageNet). These architectures optimized on proxy data are not guaranteed to be optimal on the target task. Another common way is learning with a smaller candidate set, which may require expert knowledge and indeed betrays the essence of NAS. In this paper, we present DA-NAS that can directly search the architecture for large-scale target tasks while allowing a large candidate set in a more efficient manner. Our method is based on an interesting observation that the learning speed for blocks in deep neural networks is related to the difficulty of recognizing distinct categories. We carefully design a progressive data adapted pruning strategy for efficient architecture search. It will quickly trim low performed blocks on a subset of target dataset) and then gradually find the best blocks on the whole target dataset. At this time, the original candidate set becomes as compact as possible, providing a faster search in the target task. Experiments on ImageNet verify the effectiveness of our approach. It is $ extbf{2}oldsymbol{ imes}$ faster than previous methods while the accuracy is currently state-of-the-art, at $ extbf{76.2%}$ under small FLOPs constraint. It supports an argument search space (i.e., more candidate blocks) to efficiently search the best-performing architecture.
Important applications such as mobile computing require reducing the computational costs of neural network inference. Ideally, applications would specify their preferred tradeoff between accuracy and speed, and the network would optimize this end-to-end, using classification error to remove parts of the network. Increasing speed can be done either during training – e.g., pruning filters – or during inference – e.g., conditionally executing a subset of the layers. We propose a single end-to-end framework that can improve inference efficiency in both settings. We use a combination of batch activation loss and classification loss, and Gumbel reparameterization to learn network structure. We train end-to-end, and the same technique supports pruning as well as conditional computation. We obtain promising experimental results for ImageNet classification with ResNet (45-52% less computation).
We propose a system comprised of fixed-topology neural networks having partially frozen weights, named SemifreddoNets. SemifreddoNets work as fully-pipelined hardware blocks that are optimized to have an efficient hardware implementation. Those blocks freeze a certain portion of the parameters at every layer and replace the corresponding multipliers with fixed scalers. Fixing the weights reduces the silicon area, logic delay, and memory requirements, leading to significant savings in cost and power consumption. Unlike traditional layer-wise freezing approaches, SemifreddoNets make a profitable trade between the cost and flexibility by having some of the weights configurable at different scales and levels of abstraction in the model. Although fixing the topology and some of the weights somewhat limits the flexibility, we argue that the efficiency benefits of this strategy outweigh the advantages of a fully configurable model for many use cases. Furthermore, our system uses repeatable blocks, therefore it has the flexibility to adjust model complexity without requiring any hardware change. The hardware implementation of SemifreddoNets provides up to an order of magnitude reduction in silicon area and power consumption as compared to their equivalent implementation on a general-purpose accelerator.
Data poisoning attacks on machine learning models have attracted much recent attention, wherein poisoning samples are injected at the training phase to achieve adversarial goals at test time. Although existing poisoning techniques prove to be effective in various scenarios, they rely on certain assumptions on the adversary knowledge and capability to ensure efficacy, which may be unrealistic in practice. This paper presents a new, practical targeted poisoning attack method on neural networks in vision domain, namely BlackCard. BlackCard possesses a set of critical properties for ensuring attacking efficacy in practice, which has never been simultaneously achieved by any existing work, including knowledge-oblivious, clean-label, and clean-test. Importantly, we show that the effectiveness of BlackCard can be intuitively guaranteed by a set of analytical reasoning and observations, through exploiting an essential characteristic of gradient-descent optimization which is pervasively adopted in DNN models. We evaluate the efficacy of BlackCard for generating targeted poisoning attacks via extensive experiments using various datasets and DNN models. Results show that BlackCard is effective with a rather high success rate while preserving all the claimed properties.
Deep neural networks have achieved state-of-the art performance on various computer vision tasks. However, their deployment on resource-constrained devices has been hindered due to their high computational and storage complexity. While various complexity reduction techniques, such as lightweight network architecture design and parameter quantization, have been successful in reducing the cost of implementing these networks, these methods have often been considered orthogonal. In reality, existing quantization techniques fail to replicate their success on lightweight architectures such as MobileNet. To this end, we present a novel fully differentiable non-uniform quantizer that can be seamlessly mapped onto efficient ternary-based dot product engines. We conduct comprehensive experiments on CIFAR-10, ImageNet, and Visual Wake Words datasets. The proposed quantizer (DBQ) successfully tackles the daunting task of aggressively quantizing lightweight networks such as MobileNetV1, MobileNetV2, and ShuffleNetV2. DBQ achieves state-of-the art results with minimal training overhead and provides the best (pareto-optimal) accuracy-complexity trade-off.
Deep neural networks have achieved remarkable successes in learning feature representations for visual classification. However, deep features learned by the softmax cross-entropy loss generally show excessive intra-class variations. We argue that, because the traditional softmax losses aim to optimize only the relative differences between intra-class and inter-class distances (logits), it cannot obtain representative class prototypes (class weights/centers) to regularize intra-class distances, even when the training is converged. Previous efforts mitigate this problem by introducing auxiliary regularization losses. But these modified losses mainly focus on optimizing intra-class compactness, while ignoring keeping reasonable relations between different class prototypes. These lead to weak models and eventually limit their performance. To address this problem, this paper introduces a novel Radial Basis Function (RBF) distances to replace the commonly used inner products in the softmax loss function, such that it can adaptively assign losses to regularize the intra-class and inter-class distances by reshaping the relative differences, and thus creating more representative prototypes of classes to improve optimization. The proposed RBF-Softmax loss function not only effectively reduces intra-class distances, stabilizes the training behavior, and reserves ideal relations between prototypes, but also significantly improves the testing performance. Experiments on visual recognition benchmarks including MNIST, CIFAR-10/100, and ImageNet demonstrate that the proposed RBF-Softmax achieves better results than cross-entropy and other state-of-the-art classification losses. The code is at https://github.com/2han9x1a0release/RBF-Softmax.
Many object pose estimation algorithms rely on the analysis-by-synthesis framework which requires explicit representations of individual object instances. In this paper we combine a gradient-based fitting procedure with a parametric neural image synthesis module that is capable of implicitly representing the appearance, shape and pose of entire object categories, thus rendering the need for explicit CAD models per object instance unnecessary. The image synthesis network is designed to efficiently span the pose configuration space so that model capacity can be used to capture the shape and local appearance (i.e., texture) variations jointly. At inference time the synthesized images are compared to the target via an appearance based loss and the error signal is backpropagated through the network to the input parameters, i.e. poses and appearances. Keeping the network parameters fixed, this allows for iterative optimization of the object pose, shape and appearance in a joint manner and we experimentally show that the method can recover orientation of objects with high accuracy from 2D images alone. When provided with depth measurements, to overcome scale ambiguities, the method can accurately recover the full 6DOF pose successfully.
This paper proposes a weakly-supervised learning framework for dynamics estimation from human motion. Although there are many solutions to capture pure human motion readily available, their data is not sufficient to analyze quality and efficiency of movements. Instead, the forces and moments driving human motion (the dynamics) need to be considered. Since recording dynamics is a laborious task that requires expensive sensors and complex, time-consuming optimization, dynamics data sets are small compared to human motion data sets and are rarely made public. The proposed approach takes advantage of easily obtainable motion data which enables weakly-supervised learning on small dynamics sets and weakly-supervised domain transfer. Our method includes novel neural network (NN) layers for forward and inverse dynamics during end-to-end training. On this basis, a cyclic loss between pure motion data can be minimized, i.e. no ground truth forces and moments are required during training. The proposed method achieves state-of-the-art results in terms of ground reaction force, ground reaction moment and joint torque regression and is able to maintain good performance on substantially reduced sets.
Low-bit Deep Neural Networks (low-bit DNNs) have recently received significant attention for their high efficiency. However, low-bit DNNs are often difficult to optimize due to the the saddle points in loss surfaces. Here we introduce a novel feature-based knowledge transfer framework, which utilizes a 32-bit DNN to guide the training of a low-bit DNN via feature maps. It is challenge because feature maps from two branches lie in continuous and discrete space respectively, and such mismatch has not been handled properly by existing feature transfer frameworks. In this paper, we propose to directly transfer information-rich continuous-space feature to the low-bit branch. To alleviate the negative impacts brought by the feature quantizer during the transfer process, we make two branches interact via centered cosine distance rather than the widely-used p-norms. Extensive experiments are conducted on Cifar10/100 and ImageNet. Compared with low-bit models trained directly, the proposed framework brings 0.5% to 3.4% accuracy gains to three different quantization schemes. Besides, the proposed framework can also be combined with other techniques, e.g. logits transfer, for further enhacement.
Adversarial attacks of deep neural networks have been intensively studied on image, audio, natural language, patch, and pixel classification tasks. Nevertheless, as a typical, while important real-world application, the adversarial attacks of online video object tracking that traces an object's moving trajectory instead of its category are rarely explored. In this paper, we identify a new task for the adversarial attack to visual tracking: online generating imperceptible perturbations that mislead trackers along with an incorrect (Untargeted Attack, UA) or specified trajectory (Targeted Attack, TA). To this end, we first propose a spatial-aware basic attack by adapting existing attack methods, \ie, FGSM, BIM, and C\&W, and comprehensively analyze the attacking performance. We identify that online object tracking poses two new challenges: 1) it is difficult to generate imperceptible perturbations that can transfer across frames, and 2) real-time trackers require the attack to satisfy a certain level of efficiency. To address these challenges, we further propose the spatial-aware online incremental attack (a.k.a. SPARK) that performs spatial-temporal sparse incremental perturbations online and makes the adversarial attack less perceptible. In addition, as an optimization-based method, SPARK quickly converges to very small losses within several iterations by considering historical incremental perturbations, making it much more efficient than basic attacks. The in-depth evaluation of state-of-the-art trackers (i.e., SiamRPN++ with AlexNet, MobileNetv2, and ResNet-50, and SiamDW) on OTB100, VOT2018, UAV123, and LaSOT demonstrates the effectiveness and transferability of SPARK in misleading the trackers under both UA and TA with minor perturbations.
An End-to-End OCR Text Re-organization Sequence Learning for Rich-text Detail Image Comprehension
PDF ↗Nowadays rich description on detail images help users know more about the commodities. With the help of OCR technology, the description text can be detected and recognized as auxiliary information to remove the comprehending barriers among the visual impaired users. However, for lack of proper logical structure among these OCR text blocks, it is challenging to comprehend the detail images accurately. To tackle the above problems, we propose a novel end-to-end OCR text reorganizing model. Specifically, we create a Graph Neural Network with an attention map to encode the text blocks with visual layout features, with which an attention based sequence decoder inspired by the Pointer Network and a Sinkhorn global optimization will reorder the OCR text into a proper sequence. Experimental results illustrate that our model outperforms the other baselines, and the real experiment of the blind users' experience shows that our model improves their comprehension.