Normalization layers play an important role in deep network training. As one of the most popular normalization techniques, batch normalization (BN) has shown its effectiveness in accelerating the model training speed and improving model generalization capability. The success of BN has been explained from different views, such as reducing internal covariate shift, allowing the use of large learning rate, smoothing optimization landscape, etc. To make a deeper understanding of BN, in this work we prove that BN actually introduces a certain level of noise into the sample mean and variance during the training process, while the noise level depends only on the batch size. Such a noise generation mechanism of BN regularizes the training process, and we present an explicit regularizer formulation of BN. Since the regularization strength of BN is determined by the batch size, a small batch size may cause the under-fitting problem, resulting in a less effective model. To reduce the dependency of BN on batch size, we propose a momentum BN (MBN) scheme by averaging the mean and variance of current mini-batch with the historical means and variances. With a dynamic momentum parameter, we can automatically control the noise level in the training process. As a result, MBN works very well even when the batch size is very small (e.g., 2), which is hard to achieve by traditional BN.
论文检索
输入标题、作者或关键词，从 1,358 篇学术成果中精准定位
The ubiquity of smartphone cameras has led to more and more documents being captured by cameras rather than scanned. Unlike flatbed scanners, photographed documents are often folded and crumpled, resulting in large local variance in text structure. The problem of document rectification is fundamental to the Optical Character Recognition (OCR) process on documents, and its ability to overcome geometric distortions significantly affects recognition accuracy. Despite the great progress in recent OCR systems, most still rely on a pre-process that ensures the text lines are straight and axis aligned. Recent works have tackled the problem of rectifying document images taken in-the-wild using various supervision signals and alignment means. However, they focused on global features that can be extracted from the document's boundaries, ignoring various signals that could be obtained from the document's content. We present CREASE: Content Aware Rectification using Angle Supervision, the first learned method for document rectification that relies on the document's content, the location of the words and specifically their orientation, as hints to assist in the rectification process. We utilize a novel pixel-wise angle regression approach and a curvature estimation side-task for optimizing our rectification model. Our method surpasses previous approaches in terms of OCR accuracy, geometric error and visual similarity.
Informative features play a crucial role in the single image super-resolution task. Channel attention has been demonstrated to be effective for preserving information-rich features in each layer. However, channel attention treats each convolution layer as a separate process, which is kind of missing the correlation among different layers. To address this problem, we propose a new holistic attention network (HAN), which consists of a layer attention module (LAM) and a channel-spatial attention module (CSAM), to model the holistic interdependencies among layers, channels, and positions. Specifically, the proposed LAM adaptively emphasizes hierarchical features by considering correlations among layers. Meanwhile, CSAM learns the confidence at all the positions of each channel to selectively capture more informative features. Extensive experiments demonstrate that the proposed HAN performs favorably against the state-of-the-art single image super-resolution approaches.
The task of searching certain people in videos has seen increasing potential in real-world applications, such as video organization and editing. Most existing approaches are devised to work in an offline manner, where identifies can only be inferred after an entire video is examined. This working manner precludes such methods from being applied to online services or those applications that require real-time responses. In this paper, we propose an online person search framework, which can recognize people in a video on the fly. This framework maintains a multi-modal memory bank at its heart as the basis for person recognition, and updates it dynamically with a policy obtained by reinforcement learning. Our experiments on a large movie dataset show that the proposed method is effective, not only achieving remarkable improvements over strong online schemes but also outperforming offline methods.
Online knowledge distillation has attracted increasing interest recently, which jointly learns teacher and student models or an ensemble of student models simultaneously and collaboratively. On the other hand, existing works focus more on outcome-driven learning according to knowledge like classification probabilities whereas the distilling processes which capture rich and useful intermediate features and information are largely neglected. In this work, we propose an innovative adversarial-based mutual learning network (AMLN) that introduces process-driven learning beyond outcome-driven learning for augmented online knowledge distillation. A block-wise training module is designed which guides the information flow and mutual learning among peer networks adversarially throughout different learning stages, and this spreads until the final network layer which captures more high-level information. AMLN has been evaluated under a variety of network architectures over three widely used benchmark datasets. Extensive experiments show that AMLN achieves superior performance consistently against state-of-the-art knowledge transfer methods.
Rethinking the Distribution Gap of Person Re-identification with Camera-based Batch Normalization
PDF ↗The fundamental difficulty in person re-identification (ReID) lies in learning the correspondence among individual cameras. It strongly demands costly inter-camera annotations, yet the trained models are not guaranteed to transfer well to previously unseen cameras. These problems significantly limit the application of ReID. This paper rethinks the working mechanism of conventional ReID approaches and puts forward a new solution. With an effective operator named Camera-based Batch Normalization (CBN), we force the image data of all cameras to fall onto the same subspace, so that the distribution gap between any camera pair is largely shrunk. This alignment brings two benefits. First, the trained model enjoys better abilities to generalize across scenarios with unseen cameras as well as transfer across multiple training sets. Second, we can rely on intra-camera annotations, which have been undervalued before due to the lack of cross-camera information, to achieve competitive ReID performance. Experiments on a wide range of ReID tasks demonstrate the effectiveness of our approach. The code is available at https://github.com/automan000/Camera-based-Person-ReID.
3D hand pose estimation based on RGB images has been studied for a long time. Most of the studies, however, have performed frame-by-frame estimation based on independent static images. In this paper, we attempt to not only consider the appearance of a hand but incorporate the temporal movement information of a hand in motion into the learning framework for better 3D hand pose estimation performance, which leads to the necessity of a large scale dataset with sequential RGB hand images. We propose a novel method that generates a synthetic dataset that mimics natural human hand movements by re-engineering annotations of an extant static hand pose dataset into pose-flows. With the generated dataset, we train a newly proposed recurrent framework, exploiting visuo-temporal features from sequential images of synthetic hands in motion and emphasizing temporal smoothness of estimations with a temporal consistency constraint. Our novel training strategy of detaching the recurrent layer of the framework during domain finetuning from synthetic to real allows preservation of the visuo-temporal features learned from sequential synthetic hand images. Hand poses that are sequentially estimated consequently produce natural and smooth hand movements which lead to more robust estimations. We show that utilizing temporal information for 3D hand pose estimation significantly enhances general pose estimations by outperforming state-of-the-art methods in experiments on hand pose estimation benchmarks.
We propose a weakly-supervised cell tracking method that can train a convolutional neural network (CNN) by using only the annotation of ""cell detection"" (i.e., the coordinates of cell positions) without association information, in which cell positions can be easily obtained by nuclear staining. First, we train a co-detection CNN that detects cells in successive frames by using weak-labels. Our key assumption is that the co-detection CNN implicitly learns association in addition to detection. To obtain the association information, we propose a backward-and-forward propagation method that analyzes the correspondence of cell positions in the detection maps output of the co-detection CNN. Experiments demonstrated that the proposed method can match positions by analyzing the co-detection CNN. Even though the method uses only weak supervision, the performance of our method was almost the same as the state-of-the-art supervised method.
As realistic facial manipulation technologies have achieved remarkable progress, social concerns about potential malicious abuse of these technologies bring out an emerging research topic of face forgery detection. However, it is extremely challenging since recent advances are able to forge faces beyond the perception ability of human eyes, especially in compressed images and videos. We find that mining forgery patterns with the awareness of frequency could be a cure, as frequency provides a complementary viewpoint where either subtle forgery artifacts or compression errors could be well described. To introduce frequency into the face forgery detection, we propose a novel Frequency in Face Forgery Network (F$^3$-Net), taking advantages of two different but complementary frequency-aware clues, 1) frequency-aware decomposed image components, and 2) local frequency statistics, to deeply mine the forgery patterns via our two-stream collaborative learning framework. We apply DCT as the applied frequency-domain transformation. Through comprehensive studies, we show that the proposed F$^3$-Net significantly outperforms competing state-of-the-art methods on all compression qualities in the challenging FaceForensics++ dataset, especially wins a big lead upon low-quality media.
As facial interaction systems are prevalently deployed, security and reliability of these systems become a critical issue, with substantial research efforts devoted. Among them, face anti-spoofing emerges as an important area, whose objective is to identify whether a presented face is live or spoof. Though promising progress has been achieved, existing works still have difficulty in handling complex spoof attacks and generalizing to real-world scenarios. The main reason is that current face anti-spoofing datasets are limited in both quantity and diversity. To overcome these obstacles, we contribute a large-scale face anti-spoofing dataset, extbf{CelebA-Spoof}, with the following appealing properties: extit{1) Quantity:} CelebA-Spoof comprises of 625,537 pictures of 10,177 subjects, significantly larger than the existing datasets. extit{2) Diversity:} The spoof images are captured from 8 scenes (2 environments all_papers.txt decode_tex_noligatures.sh decode_tex_noligatures.sh~ decode_tex.sh decode_tex.sh~ ECCV_abstracts.csv ECCV_abstracts_good.csv ECCV_abstracts_good.csv.old ECCV_abstracts_ori.csv ECCV.csv ECCV.csv~ ECCV_new.csv generate_list.sh generate_list.sh~ generate_overview.sh generate_overview.sh~ gen.sh pdflist pdflist.copied RCS snippet.html 4 illumination conditions) with more than 10 sensors. extit{3) Annotation Richness:} CelebA-Spoof contains 10 spoof type annotations, as well as the 40 attribute annotations inherited from the original CelebA dataset. Equipped with CelebA-Spoof, we carefully benchmark existing methods in a unified multi-task framework, extbf{Auxiliary Information Embedding Network (AENet)}, and reveal several valuable observations. Our key insight is that, compared with the commonly-used binary supervision or mid-level geometric representations, rich semantic annotations as auxiliary tasks can greatly boost the performance and generalizability of face anti-spoofing across a wide range of spoof attacks. Through comprehensive studies, we show that CelebA-Spoof serves as an effective training data source. Models trained on CelebA-Spoof (without fine-tuning) exhibit state-of-the-art performance on standard benchmarks such as CASIA-MFSD. The datasets are available at \href{https://github.com/Davidzhangyuanhan/CelebA-Spoof}{https://github.com/Davidzhangyuanhan/CelebA-Spoof} .
Stereophonic audio is an indispensable ingredient to enhance human auditory experience. Recent research has explored the usage of visual information as guidance to generate binaural or ambisonic audio from mono ones with stereo supervision. However, this fully supervised paradigm suffers from an inherent drawback: the recording of stereophonic audio usually requires delicate devices that are expensive for wide accessibility. To overcome this challenge, we propose to leverage the vastly available mono data to facilitate the generation of stereophonic audio. Our key observation is that the task of visually indicated audio separation also maps independent audios to their corresponding visual positions, which shares a similar objective with stereophonic audio generation. We integrate both stereo generation and source separation into a unified framework, Sep-Stereo, by considering source separation as a particular type of audio spatialization. Specifically, a novel associative pyramid network architecture is carefully designed for audio-visual feature fusion. Extensive experiments demonstrate that our framework can improve the stereophonic audio generation results while performing accurate sound separation with a shared backbone.
We present a method that infers spatial arrangements and shapes of humans and objects in a globally consistent 3D scene, all from a single image in-the-wild captured in an uncontrolled environment. Notably, our method runs on datasets without any scene- or object-level 3D supervision. Our key insight is that considering humans and objects jointly gives rise to ``3D common sense"" constraints that can be used to resolve ambiguity. In particular, we introduce a scale loss that learns the distribution of object size from data; an occlusion-aware silhouette re-projection loss to optimize object pose; and a human-object interaction loss to capture the spatial layout of objects with which humans interact. We empirically validate that our constraints dramatically reduce the space of likely 3D spatial configurations. We demonstrate our approach on challenging, in-the-wild images of humans interacting with large objects (such as bicycles, motorcycles, and surfboards) and handheld objects (such as laptops, tennis rackets, and skateboards). We quantify the ability of our approach to recover human-object arrangements and outline remaining challenges in this relatively unexplored domain. The project webpage can be found at https://jasonyzhang.com/phosa.
Sufficient knowledge extraction from the teacher network plays a critical role in the knowledge distillation task to improve the performance of the student network. Existing methods mainly focus on the consistency of instance-level features and their relationships, but neglect the local features and their correlation, which also contain many details and discriminative patterns. In this paper, we propose the local correlation exploration framework for knowledge distillation. It models three kinds of local knowledge, including intra-instance local relationship, inter-instance relationship on the same local position, and the inter-instance relationship across different local positions. Moreover, to make the student focus on those informative local regions of the teacher's feature maps, we propose a novel class-aware attention module to highlight the class-relevant regions and remove the confusing class-irrelevant regions, which makes the local correlation knowledge more accurate and valuable. We conduct extensive experiments and ablation studies on challenging datasets, including CIFAR100 and ImageNet, to show our superiority over the state-of-the-art methods.
Neural network quantization is an effective way to compress deep models and improve their execution latency and energy efficiency, so that they can be deployed on mobile or embedded devices. Existingquantization methods require original data for calibration or fine-tuning to get better performance. However, in many real-world scenarios, the data may not be available due to confidential or private issues, thereby making existing quantization methods not applicable. Moreover, due to the absence of original data, the recently developed generative adversarial networks (GANs) cannot be applied to generate data. Although the full-precision model may contain rich data information, such information alone is hard to exploit for recovering the original data or generating new meaningful data. In this paper, we investigate a simple-yet-effective method called Generative Low-bitwidth Data Free Quantization(GDFQ) to remove the data dependence burden. Specifically, we propose a knowledge matching generator to produce meaningful fake data by exploiting classification boundary knowledge and distribution information in the pre-trained model. With the help of generated data, we can quantize a model by learning knowledge from the pre-trained model. Extensive experiments on three data sets demonstrate the effectiveness of our method. More critically, our method achieves much higher accuracy on4-bit quantization than the existing data free quantization method. Code is available at https://github.com/xushoukai/GDFQ.
Regional Homogeneity: Towards Learning Transferable Universal Adversarial Perturbations Against Defenses
PDF ↗This paper focuses on learning transferable adversarial examples specifically against defense models (models to defense adversarial attacks). In particular, we show that a simple universal perturbation can fool a series of state-of-the-art defenses.Adversarial examples generated by existing attacks are generally hard to transfer to defense models. We observe the property of regional homogeneity in adversarial perturbations and suggest that the defenses are less robust to regionally homogeneous perturbations. Therefore, we propose an effective transforming paradigm and a customized gradient transformer module to transform existing perturbations into regionally homogeneous ones. Without explicitly forcing the perturbations to be universal, we observe that a well-trained gradient transformer module tends to output input-independent gradients (hence universal) benefiting from the under-fitting phenomenon. Thorough experiments demonstrate that our work significantly outperforms the prior art attacking algorithms (either image-dependent or universal ones) by an average improvement of 14.0% when attacking 9 defenses in the transfer-based attack setting. In addition to the cross-model transferability, we also verify that regionally homogeneous perturbations can well transfer across different vision tasks (attacking with the semantic segmentation task and testing on the object detection task). The code is available here: https://github.com/LiYingwei/Regional-Homogeneity.
Humans view the world through many sensory channels, e.g., the long-wavelength light channel, viewed by the left eye, or the high-frequency vibrations channel, heard by the right ear. Each view is noisy and incomplete, but important factors, such as physics, geometry, and semantics, tend to be shared between all views (e.g., a ``dog"" can be seen, heard, and felt). We investigate the classic hypothesis that a powerful representation is one that models view-invariant factors. We study this hypothesis under the framework of multiview contrastive learning, where we learn a representation that aims to maximize mutual information between different views of the same scene but is otherwise compact. Our approach scales to any number of views, and is view-agnostic. We analyze key properties of the approach that make it work, finding that the contrastive loss outperforms a popular alternative based on cross-view prediction, and that the more views we learn from, the better the resulting representation captures underlying scene semantics.
In this paper, we introduce Foley Music, a system that can synthesize plausible music for a silent video clip about people playing musical instruments. We first identify two key intermediate representations for a successful video to music generator: body keypoints from videos and MIDI events from audio recordings. We then formulate music generation from videos as a motion-to-midi translation problem. We present a Graph-Transformer framework that can accurately predict MIDI event sequences in accordance with the body movements; The MIDI event can then be converted to realistic music using an off-the-shelf music synthesizer tool. We demonstrate the effectiveness of our models on videos containing a variety of music performances. Experimental results show that our model outperforms several existing systems in generating music that is pleasant to listen to. More importantly, the MIDI representations are fully interpretable and transparent, thus enabling us to perform music editing flexibly. We encourage the readers to watch the supplementary video with audio turned on to experience the results.
Margin-based deep face recognition methods (e.g. SphereFace, CosFace, and ArcFace) have achieved remarkable success in unconstrained face recognition. However, these methods are susceptible to the massive label noise in the training data and thus require laborious human effort to clean the datasets. In this paper, we relax the intra-class constraint of ArcFace to improve the robustness to label noise. More specifically, we design $K$ sub-centers for each class and the training sample only needs to be close to any of the $K$ positive sub-centers instead of the only one positive center. The proposed sub-center ArcFace encourages one dominant sub-class that contains the majority of clean faces and non-dominant sub-classes that include hard or noisy faces. Extensive experiments confirm the robustness of sub-center ArcFace under massive real-world noise. After the model achieves enough discriminative power, we directly drop non-dominant sub-centers and high-confident noisy samples, which helps recapture intra-compactness, decrease the influence from noise, and achieve comparable performance compared to ArcFace trained on the manually cleaned dataset. By taking advantage of the large-scale raw web faces (Celeb500K), sub-center Arcface achieves state-of-the-art performance on IJB-B, IJB-C, MegaFace, and FRVT.
Local features such as SIFT and its affine and learned variants provide region-to-region rather than point-to-point correspondences. It has recently been exploited to create new minimal solvers for classical problems such as homography, essential and fundamental matrix estimation. The main argument in favor of such solvers is that their minimal sample size is smaller, e.g., only two instead of four matches are required to estimate a homography. Works proposing such solvers often claim a significant improvement in run-time thanks to fewer RANSAC iterations. We show that this argument is not valid in practice if the solvers are used naively as noise in the local feature geometries often causes RANSAC to fail to find an accurate solution. To overcome this, we propose guidelines for effective use of region-to-region matches in the course of a full model estimation pipeline. We propose a method for refining the local feature geometries by symmetric intensity-based matching, combine uncertainty propagation inside RANSAC with preemptive model verification, show a general scheme for computing uncertainty of minimal solvers results, and adapt the sample cheirality check for homography estimation to region-to-region correspondences. Our experiments show that affine solvers can achieve accuracy comparable to point-based solvers at faster run-times when following our guidelines.
Recent end-to-end trainable methods for scene text spotting, integrating detection and recognition, showed much progress. However, most of the current arbitrary-shape scene text spotters use region proposal networks (RPN) to produce proposals. RPN relies heavily on manually designed anchors and its proposals are represented with axis-aligned rectangles. The former presents difficulties in handling text instances of extreme aspect ratios or irregular shapes, and the latter often includes multiple neighboring instances into a single proposal, in cases of densely oriented text. To tackle these problems, we propose Mask TextSpotter v3, an end-to-end trainable scene text spotter that adopts a Segmentation Proposal Network (SPN) instead of an RPN. Our SPN is anchor-free and gives accurate representations of arbitrary-shape proposals. It is therefore superior to RPN in detecting text instances of extreme aspect ratios or irregular shapes. Furthermore, the accurate proposals produced by SPN allow masked RoI features to be used for decoupling neighboring text instances. As a result, our Mask TextSpotter v3 can handle text instances of extreme aspect ratios or irregular shapes, and its recognition accuracy won't be affected by nearby text or background noise. Specifically, we outperform state-of-the-art methods by 21.9 percent on the Rotated ICDAR 2013 dataset (rotation robustness), 5.9 percent on the Total-Text dataset (shape robustness), and achieve state-of-the-art performance on the MSRA-TD500 dataset (aspect ratio robustness). Code is available at: https://github.com/MhLiao/MaskTextSpotterV3