Recent methods for visual question answering rely on large-scale annotated datasets. Manual annotation of questions and answers for videos, however, is tedious, expensive and prevents scalability. In this work, we propose to avoid manual annotation and generate a large-scale training dataset for video question answering making use of automatic cross-modal supervision. We leverage a question generation transformer trained on text data and use it to generate question-answer pairs from transcribed video narrations. Given narrated videos, we then automatically generate the HowToVQA69M dataset with 69M video-question-answer triplets. To handle the open vocabulary of diverse answers in this dataset, we propose a training procedure based on a contrastive loss between a video-question multi-modal transformer and an answer transformer. We introduce the zero-shot VideoQA task and show excellent results, in particular for rare answers. Furthermore, we demonstrate our method to significantly outperform the state of the art on MSRVTT-QA, MSVD-QA, ActivityNet-QA and How2QA. Finally, for a detailed evaluation we introduce iVQA, a new VideoQA dataset with reduced language biases and high-quality redundant manual annotations.
论文检索
输入标题、作者或关键词，从 1,612 篇学术成果中精准定位
Recent studies have shown that many deep metric learning loss functions perform very similarly under the same experimental conditions. One potential reason for this unexpected result is that all losses let the network focus on similar image regions or properties. In this paper, we investigate this by conducting a two-step analysis to extract and compare the learned visual features of the same model architecture trained with different loss functions: First, we compare the learned features on the pixel level by correlating saliency maps of the same input images. Second, we compare the clustering of embeddings for several image properties, e.g. object color or illumination. To provide independent control over these properties, photo-realistic 3D car renders similar to images in the Cars196 dataset are generated. In our analysis, we compare 14 pretrained models from a recent study and find that, even though all models perform similarly, different loss functions can guide the model to learn different features. We especially find differences between classification and ranking based losses. Our analysis also shows that some seemingly irrelevant properties can have significant influence on the resulting embedding. We encourage researchers from the deep metric learning community to use our methods to get insights into the features learned by their proposed methods.
How To Design a Three-Stage Architecture for Audio-Visual Active Speaker Detection in the Wild
PDF ↗Successful active speaker detection requires a three-stage pipeline: (i) audio-visual encoding for all speakers in the clip, (ii) inter-speaker relation modeling between a reference speaker and the background speakers within each frame, and (iii) temporal modeling for the reference speaker. Each stage of this pipeline plays an important role for the final performance of the created architecture. Based on a series of controlled experiments, this work presents several practical guidelines for audio-visual active speaker detection. Correspondingly, we present a new architecture called ASDNet, which achieves a new state-of-the-art on the AVA-ActiveSpeaker dataset with a mAP of 93.5% outperforming the second best with a large margin of 4.7%. Our code and pretrained models are publicly available.
We present RePOSE, a fast iterative refinement method for 6D object pose estimation. Prior methods perform refinement by feeding zoomed-in input and rendered RGB images into a CNN and directly regressing an update of a refined pose. Their runtime is slow due to the computational cost of CNN, which is especially prominent in multiple-object pose refinement. To overcome this problem, RePOSE leverages image rendering for fast feature extraction using a 3D model with a learnable texture. We call this deep texture rendering, which uses a shallow multi-layer perceptron to directly regress a view-invariant image representation of an object. Furthermore, we utilize differentiable Levenberg-Marquard (LM) optimization to refine a pose fast and accurately by minimizing the feature-metric error between the input and rendered image representations without the need of zooming in. These image representations are trained such that differentiable LM optimization converges within few iterations. Consequently, RePOSE runs at 92 FPS and achieves state-of-the-art accuracy of 51.6% on the Occlusion LineMOD dataset - a 4.1% absolute improvement over the prior art, and comparable result on the YCB-Video dataset with a much faster runtime. The code is available at https://github.com/sh8/repose.
Despite significant progress, we show that state of the art 3D human pose and shape estimation methods remain sensitive to partial occlusion and can produce dramatically wrong predictions although much of the body is observable. To address this, we introduce a soft attention mechanism, called the Part Attention REgressor (PARE), that learns to predict body-part-guided attention masks. We observe that state-of-the-art methods rely on global feature representations, making them sensitive to even small occlusions. In contrast, PARE's part-guided attention mechanism overcomes these issues by exploiting information about the visibility of individual body parts while leveraging information from neighboring body-parts to predict occluded parts. We show qualitatively that PARE learns sensible attention masks, and quantitative evaluation confirms that PARE achieves more accurate and robust reconstruction results than existing approaches on both occlusion-specific and standard benchmarks. The code and data are available for research purposes at https://pare.is.tue.mpg.de/
We investigate the impact of aliasing on generalization in Deep Convolutional Networks and show that data augmentation schemes alone are unable to prevent it due to structural limitations in widely used architectures. Drawing insights from frequency analysis theory, we take a closer look at Resnet and EfficientNet architectures and review the trade-off between aliasing and information loss in each of their major components. We show how to mitigate aliasing by inserting non-trainable low-pass filters at key locations, particularly where networks lack the capacity to learn them. These simple architectural changes lead to substantial improvements in generalization on i.i.d. and even more on out-of-distribution conditions, such as image classification under natural corruptions on ImageNet-C and few-shot learning on Meta-Dataset. State-of-the art results are achieved on both datasets without introducing additional trainable parameters and using the default hyper-parameters of open source codebases.
In this paper, we present Co-scale conv-attentional image Transformers (CoaT), a Transformer-based image classifier equipped with co-scale and conv-attentional mechanisms. First, the co-scale mechanism maintains the integrity of Transformers' encoder branches at individual scales, while allowing representations learned at different scales to effectively communicate with each other; we design a series of serial and parallel blocks to realize the co-scale mechanism. Second, we devise a conv-attentional mechanism by realizing a relative position embedding formulation in the factorized attention module with an efficient convolution-like implementation. CoaT empowers image Transformers with enriched multi-scale and contextual modeling capabilities. On ImageNet, relatively small CoaT models attain superior classification results compared with similar-sized convolutional neural networks and image/vision Transformers. The effectiveness of CoaT's backbone is also illustrated on object detection and instance segmentation, demonstrating its applicability to downstream computer vision tasks.
Neural Architecture Search (NAS) can automatically design well-performed architectures of Deep Neural Networks (DNNs) for the tasks at hand. However, one bottleneck of NAS is the prohibitively computational cost largely due to the expensive performance evaluation. The neural predictors can directly estimate the performance without any training of the DNNs to be evaluated, thus have drawn increasing attention from researchers. Despite their popularity, they also suffer a severe limitation: the shortage of annotated DNN architectures for effectively training the neural predictors. In this paper, we proposed Homogeneous Architecture Augmentation for Neural Predictor (HAAP) of DNN architectures to address the issue aforementioned. Specifically, a homogeneous architecture augmentation algorithm is proposed in HAAP to generate sufficient training data taking the use of homogeneous representation. Furthermore, the one-hot encoding strategy is introduced into HAAP to make the representation of DNN architectures more effective. The experiments have been conducted on both NAS-Benchmark-101 and NAS-Bench-201 dataset. The experimental results demonstrate that the proposed HAAP algorithm outperforms the state of the arts compared, yet with much less training data. In addition, the ablation studies on both benchmark datasets have also shown the universality of the homogeneous architecture augmentation. Our code has been made available at https://github.com/lyq998/HAAP.
Explainability for machine learning models has gained considerable attention within the research community given the importance of deploying more reliable machine-learning systems. In computer vision applications, generative counterfactual methods indicate how to perturb a model's input to change its prediction, providing details about the model's decision-making. Current methods tend to generate trivial counterfactuals about a model's decisions, as they often suggest to exaggerate or remove the presence of the attribute being classified. For the machine learning practitioner, these types of counterfactuals offer little value, since they provide no new information about undesired model or data biases. In this work, we identify the problem of trivial counterfactual generation and we propose DiVE to alleviate it. DiVE learns a perturbation in a disentangled latent space that is constrained using a diversity-enforcing loss to uncover multiple valuable explanations about the model's prediction. Further, we introduce a mechanism to prevent the model from producing trivial explanations. Experiments on CelebA and Synbols demonstrate that our model improves the success rate of producing high-quality valuable explanations when compared to previous state-of-the-art methods.
State-of-the-art approaches for visually-guided audio source separation typically assume sources that have characteristic sounds, such as musical instruments. These approaches often ignore the visual context of these sound sources or avoid modeling object interactions that may be useful to better characterize the sources, especially when the same object class may produce varied sounds from distinct interactions. To address this challenging problem, we propose Audio Visual Scene Graph Segmenter (AVSGS), a novel deep learning model that embeds the visual structure of the scene as a graph and segments this graph into subgraphs, each subgraph being associated with a unique sound obtained by co-segmenting the audio spectrogram. At its core, AVSGS uses a recursive neural network that emits mutually-orthogonal sub-graph embeddings of the visual graph using multi-head attention. These embeddings are used for conditioning an audio encoder-decoder towards source separation. Our pipeline is trained end-to-end via a self-supervised task consisting of separating audio sources using the visual graph from artificially mixed sounds. In this paper, we also introduce an ""in the wild" video dataset for sound source separation that contains multiple non-musical sources, which we call Audio Separation in the Wild (ASIW). This dataset is adapted from the AudioCaps dataset, and provides a challenging, natural, and daily-life setting for source separation. Thorough experiments on the proposed ASIW and the standard MUSIC datasets demonstrate state-of-the-art sound separation performance of our method against recent prior approaches.
Compact convolutional neural networks (CNNs) have witnessed exceptional improvements in performance in recent years. However, they still fail to provide the same predictive power as CNNs with a large number of parameters. The diverse and even abundant features captured by the layers is an important characteristic of these successful CNNs. However, differences in this characteristic between large CNNs and their compact counterparts have rarely been investigated. In compact CNNs, due to the limited number of parameters, abundant features are unlikely to be obtained, and feature diversity becomes an essential characteristic. Diverse features present in the activation maps derived from a data point during model inference may indicate the presence of a set of unique descriptors necessary to distinguish between objects of different classes. In contrast, data points with low feature diversity may not provide a sufficient amount of unique descriptors to make a valid prediction; we refer to them as random predictions. Random predictions can negatively impact the optimization process and harm the final performance. This paper proposes addressing the problem raised by random predictions by reshaping the standard cross-entropy to make it biased toward data points with a limited number of unique descriptive features. Our novel Bias Loss focuses the training on a set of valuable data points and prevents the vast number of samples with poor learning features from misleading the optimization process. Furthermore, to show the importance of diversity, we present a family of SkipblockNet models whose architectures are brought to boost the number of unique descriptors in the last layers. Experiments conducted on benchmark datasets demonstrate the superiority of the proposed loss function over the cross-entropy loss. Moreover, our SkipblockNet-M can achieve 1% higher classification accuracy than MobileNetV3 Large with similar computational cost on the ImageNet ILSVRC-2012 classification dataset. The code is available on the link - https://github.com/lusinlu/biasloss_skipblocknet.
Efficient action recognition has become crucial to extend the success of action recognition to many real-world applications. Contrary to most existing methods, which mainly focus on selecting salient frames to reduce the computation cost, we focus more on making the most of the selected frames. To this end, we employ two networks of different capabilities that operate in tandem to efficiently recognize actions. Given a video, the lighter network processes more frames while the heavier one only processes a few. In order to enable the effective interaction between the two, we propose dynamic knowledge propagation based on a cross-attention mechanism. This is the main component of our framework that is essentially a student-teacher architecture, but as the teacher model continues to interact with the student model during inference, we call it a dynamic student-teacher framework. Through extensive experiments, we demonstrate the effectiveness of each component of our framework. Our method outperforms competing state-of-the-art methods on two video datasets: ActivityNet-v1.3 and Mini-Kinetics.
We present SemiHand, a semi-supervised framework for 3D hand pose estimation from monocular images. We pre-train the model on labelled synthetic data and fine-tune it on unlabelled real-world data by pseudo-labeling with consistency training. By design, we introduce data augmentation of differing difficulties, consistency regularizer, label correction and sample selection for RGB-based 3D hand pose estimation. In particular, by approximating the hand masks from hand poses, we propose a cross-modal consistency and leverage semantic predictions to guide the predicted poses. Meanwhile, we introduce pose registration as label correction to guarantee the biomechanical feasibility of hand bone lengths. Experiments show that our method achieves a favorable improvement on real-world datasets after fine-tuning.
Interactive retrieval for online fashion shopping provides the ability of changing image retrieval results according to the user feedback. One common problem in interactive retrieval is that a specific user interaction (e.g., changing the color of a T-shirt) causes other aspects to change inadvertently (e.g., the results have a sleeve type different from that of the query). This is a consequence of existing methods learning visual representations that are entangled in the embedding space, which limits the controllability of the retrieved results. We propose to leverage on the semantics of visual attributes to train convolutional networks that learn attribute-specific subspaces for each attribute type to obtain disentangled representations. Operations, such as swapping out a particular attribute value for another, impact the attribute at hand and leave others untouched. We show that our model can be tailored to deal with different retrieval tasks while maintaining its disentanglement property. We obtained state-of-the-art performance on three interactive fashion retrieval tasks: attribute manipulation retrieval, conditional similarity retrieval, and outfit complementary item retrieval. We will make code and models publicly available.
We present a simple yet highly generalizable method for explaining interacting parts within a neural network's reasoning process. First, we design an algorithm based on cross derivatives for computing statistical interaction effects between individual features, which is generalized to both 2-way and higher-order (3-way or more) interactions. We present results side by side with a weight-based attribution technique, corroborating that cross derivatives are a superior metric for both 2-way and higher-order interaction detection. Moreover, we extend the use of cross derivatives as an explanatory device in neural networks to the computer vision setting by expanding Grad-CAM, a popular gradient-based explanatory tool for CNNs, to the higher order. While Grad-CAM can only explain the importance of individual objects in images, our method, which we call Taylor-CAM, can explain a neural network's relational reasoning across multiple objects. We show the success of our explanations both qualitatively and quantitatively, including with a user study. We will release all code as a tool package to facilitate explainable deep learning.
Modern cameras are equipped with a wide array of sensors that enable recording the geospatial context of an image. Taking advantage of this, we explore depth estimation under the assumption that the camera is geocalibrated, a problem we refer to as geo-enabled depth estimation. Our key insight is that if capture location is known, the corresponding overhead viewpoint offers a valuable resource for understanding the scale of the scene. We propose an end-to-end architecture for depth estimation that uses geospatial context to infer a synthetic ground-level depth map from a co-located overhead image, then fuses it inside of an encoder/decoder style segmentation network. To support evaluation of our methods, we extend a recently released dataset with overhead imagery and corresponding height maps. Results demonstrate that integrating geospatial context significantly reduces error compared to baselines, both at close ranges and when evaluating at much larger distances than existing benchmarks consider.
Stereo depth estimation relies on optimal correspondence matching between pixels on epipolar lines in the left and right images to infer depth. In this work, we revisit the problem from a sequence-to-sequence correspondence perspective to replace cost volume construction with dense pixel matching using position information and attention. This approach, named STereo TRansformer (STTR), has several advantages: It 1) relaxes the limitation of a fixed disparity range, 2) identifies occluded regions and provides confidence estimates, and 3) imposes uniqueness constraints during the matching process. We report promising results on both synthetic and real-world datasets and demonstrate that STTR generalizes across different domains, even without fine-tuning.
Diagnosing diseases from medical radiographs and writing reports requires professional knowledge and is time-consuming. To address this, automatic medical report generation approaches have recently gained interest. However, identifying diseases as well as correctly predicting their corresponding sizes, locations and other medical description patterns, which is essential for generating high-quality reports, is challenging. Although previous methods focused on producing readable reports, how to accurately detect and describe findings that match with the query X-Ray has not been successfully addressed. In this paper, we propose a multi-modality semantic attention model to integrate visual features, predicted key finding embeddings, as well as clinical features, and progressively decode reports with visual-textual semantic consistency. First, multi-modality features are extracted and attended with the hidden states from the sentence decoder, to encode enriched context vectors for better decoding a report. These modalities include regional visual features of scans, semantic word embeddings of the top-K findings predicted with high probabilities, and clinical features of indications. Second, the progressive report decoder consists of a sentence decoder and a word decoder, where we propose image-sentence matching and description accuracy losses to constrain the visual-textual semantic consistency. Extensive experiments on the public MIMIC-CXR and IU X-Ray datasets show that our model achieves consistent improvements over the state-of-the-art methods.
For several emerging technologies such as augmented reality, autonomous driving and robotics, visual localization is a critical component. Directly regressing camera pose/3D scene coordinates from the input image using deep neural networks has shown great potential. However, such methods assume a stationary data distribution with all scenes simultaneously available during training. In this paper, we approach the problem of visual localization in a continual learning setup -- whereby the model is trained on scenes in an incremental manner. Our results show that similar to the classification domain, non-stationary data induces catastrophic forgetting in deep networks for visual localization. To address this issue, a strong baseline based on storing and replaying images from a fixed buffer is proposed. Furthermore, we propose a new sampling method based on coverage score (Buff-CS) that adapts the existing sampling strategies in the buffering process to the problem of visual localization. Results demonstrate consistent improvements over standard buffering methods on two challenging datasets -- 7Scenes, 12Scenes, and also 19Scenes by combining the former scenes.
Self-supervised representation learning is able to learn semantically meaningful features; however, much of its recent success relies on multiple crops of an image with very few objects. Instead of learning view-invariant representation from simple images, humans learn representations in a complex world with changing scenes by observing object movement, deformation, pose variation, and ego motion. Motivated by this ability, we present a new self-supervised learning representation framework that can be directly deployed on a video stream of complex scenes with many moving objects. Our framework features a simple flow equivariance objective that encourages the network to predict the features of another frame by applying a flow transformation to the features of the current frame. Our representations, learned from high-resolution raw video, can be readily used for downstream tasks on static images. Readout experiments on challenging semantic segmentation, instance segmentation, and object detection benchmarks show that we are able to outperform representations obtained from previous state-of-the-art methods including SimCLR and BYOL.