Compared with the domain-specific model, the vision-language pre-training models (VLPMs) have shown superior performance on downstream tasks with fast fine-tuning process. For example, ERNIE-ViL, Oscar and UNIMO trained VLPMs with a uniform transformers stack architecture and large amounts of image-text paired data, achieving remarkable results on downstream tasks such as image-text reference(IR and TR), vision question answering (VQA) and image captioning (IC) etc. During the training phase, VLPMs are always fed with a combination of multiple public datasets to meet the demand of large-scare training data. However, due to the unevenness of data distribution including size, task type and quality, using the mixture of multiple datasets for model training can be problematic. In this work, we introduce a large-scale multi-modal corpora named WuDaoMM, totally containing more than 650M image-text pairs. Specifically, about 600 million pairs of data are collected from multiple webpages in which image and caption present weak correlation, and the other 50 million strong-related image-text pairs are collected from some high-quality graphic websites. We also release a base version of WuDaoMM with 5 million strong-correlated image-text pairs, which is sufficient to support the common cross-modal model pre-training. Besides, we trained both an understanding and a generation vision-language (VL) model to test the dataset effectiveness. The results show that WuDaoMM can be applied as an efficient dataset for VLPMs, especially for the model in text-to-image generation task. The data is released at https://data.wudaoai.cn
Recently, template-based trackers have become the leading tracking algorithms with promising performance in terms of efficiency and accuracy. However, the correlation operation between query feature and the given template only exploits accurate target localization, leading to state estimation error especially when the target suffers from severe deformable variations. To address this issue, segmentation-based trackers have been proposed that employ per-pixel matching to improve the tracking performance of deformable objects effectively. However, most of existing trackers only refer to the target features in the initial frame, thereby lacking the discriminative capacity to handle challenging factors, e.g., similar distractors, background clutter, appearance change, etc. To this end, we propose a dynamic compact memory embedding to enhance the discrimination of the segmentation-based deformable visual tracking method. Specifically, we initialize a memory embedding with the target features in the first frame. During the tracking process, the current target features that have high correlation with existing memory are updated to the memory embedding online. To further improve the segmentation accuracy for deformable objects, we employ a point-to-global matching strategy to measure the correlation between the pixel-wise query features and the whole template, so as to capture more detailed deformation information. Extensive evaluations on six challenging tracking benchmarks including VOT2016, VOT2018, VOT2019, GOT-10K, TrackingNet, and LaSOT demonstrate the superiority of our method over recent remarkable trackers. Besides, our method outperforms the excellent segmentation-based trackers, i.e., D3S and SiamMask on DAVIS2017 benchmark.
Despite the success of neural dialogue systems in achieving high performance on the leader-board, they cannot meet users' requirements in practice, due to their poor reasoning skills. The underlying reason is that most neural dialogue models only capture the syntactic and semantic information, but fail to model the logical consistency between the dialogue history and the generated response. Recently, a new multi-turn dialogue reasoning task has been proposed, to facilitate dialogue reasoning research. However, this task is challenging, because there are only slight differences between the illogical response and the dialogue history. How to effectively solve this challenge is still worth exploring. This paper proposes a Fine-grained Comparison Model (FCM) to tackle this problem. Inspired by human's behavior in reading comprehension, a comparison mechanism is proposed to focus on the fine-grained differences in the representation of each response candidate. Specifically, each candidate representation is compared with the whole history to obtain a history consistency representation. Furthermore, the consistency signals between each candidate and the speaker's own history are considered to drive a model to prefer a candidate that is logically consistent with the speaker's history logic. Finally, the above consistency representations are employed to output a ranking list of the candidate responses for multi-turn dialogue reasoning. Experimental results on two public dialogue datasets show that our method obtains higher ranking scores than the baseline models.
Fine-grained location prediction on smart phones can be used to improve app/system performance. Application scenarios include video quality adaptation as a function of the 5G network quality at predicted user locations, and augmented reality apps that speed up content rendering based on predicted user locations. Such use cases require prediction error in the same range as the GPS error, and no existing works on location prediction can achieve this level of accuracy. We present a system for fine-grained location prediction (FGLP) of mobile users, based on GPS traces collected on the phones. FGLP has two components: a federated learning framework and a prediction model. The framework runs on the phones of the users and also on a server that coordinates learning from all users in the system. FGLP represents the user location data as relative points in an abstract 2D space, which enables learning across different physical spaces. The model merges Bidirectional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNN), where BiLSTM learns the speed and direction of the mobile users, and CNN learns information such as user movement preferences. FGLP uses federated learning to protect user privacy and reduce bandwidth consumption. Our experimental results, using a dataset with over 600,000 users, demonstrate that FGLP outperforms baseline models in terms of prediction accuracy. We also demonstrate that FGLP works well in conjunction with transfer learning, which enables model reusability. Finally, benchmark results on several types of Android phones demonstrate FGLP's feasibility in real life.
It is well known that adversarial attacks can fool deep neural networks with imperceptible perturbations. Although adversarial training significantly improves model robustness, failure cases of defense still broadly exist. In this work, we find that the adversarial attacks can also be vulnerable to small perturbations. Namely, on adversarially-trained models, perturbing adversarial examples with a small random noise may invalidate their misled predictions. After carefully examining state-of-the-art attacks of various kinds, we find that all these attacks have this deficiency to different extents. Enlightened by this finding, we propose to counter attacks by crafting more effective defensive perturbations. Our defensive perturbations leverage the advantage that adversarial training endows the ground-truth class with smaller local Lipschitzness. By simultaneously attacking all the classes, the misled predictions with larger Lipschitzness can be flipped into correct ones. We verify our defensive perturbation with both empirical experiments and theoretical analyses on a linear model. On CIFAR10, it boosts the state-of-the-art model from 66.16% to 72.66% against the four attacks of AutoAttack, including 71.76% to 83.30% against the Square attack. On ImageNet, the top-1 robust accuracy of FastAT is improved from 33.18% to 38.54% under the 100-step PGD attack.
Bounding box regression is an important component in object detection. Recent work has shown the promising performance by optimizing the Intersection over Union (IoU) as loss. However, IoU-based loss has the gradient vanish problem in the case of low overlapping bounding boxes, and the model could easily ignore these simple cases. In this paper, we propose Side Overlap (SO) loss by maximizing the side overlap of two bounding boxes, which puts more penalty for low overlapping bounding box cases. Besides, to speed up the convergence, the Corner Distance (CD) is added into the objective function. Combining the Side Overlap and Corner Distance, we get a new regression objective function, Side and Corner Align Loss (SCALoss). The SCALoss is well-correlated with IoU loss, which also benefits the evaluation metric but produces more penalty for low-overlapping cases. It can serve as a comprehensive similarity measure, leading the better localization performance and faster convergence speed. Experiments on COCO and PASCAL VOC benchmarks show that SCALoss can bring consistent improvement and outperform $\ell_n$ loss and IoU based loss with popular object detectors such as YOLOV3, SSD, Reppoints, Faster-RCNN.
Question Answering (QA) models over Knowledge Bases (KBs) are capable of providing more precise answers by utilizing relation information among entities. Although effective, most of these models solely rely on fixed relation representations to obtain answers for different question-related KB subgraphs. Hence, the rich structured information of these subgraphs may be overlooked by the relation representation vectors. Meanwhile, the direction information of reasoning, which has been proven effective for the answer prediction on graphs, has not been fully explored in existing work. To address these challenges, we propose a novel neural model, Relation-updated Direction-guided Answer Selector (RDAS), which converts relations in each subgraph to additional nodes to learn structure information. Additionally, we utilize direction information to enhance the reasoning ability. Experimental results show that our model yields substantial improvements on two widely used datasets.
As an instance-level recognition problem, re-identification (re-ID) requires models to capture diverse features. However, with continuous training, re-ID models pay more and more attention to the salient areas. As a result, the model may only focus on few small regions with salient representations and ignore other important information. This phenomenon leads to inferior performance, especially when models are evaluated on small inter-identity variation data. In this paper, we propose a novel network, Erasing-Salient Net (ES-Net), to learn comprehensive features by erasing the salient areas in an image. ES-Net proposes a novel method to locate the salient areas by the confidence of objects and erases them efficiently in a training batch. Meanwhile, to mitigate the over-erasing problem, this paper uses a trainable pooling layer P-pooling that generalizes global max and global average pooling. Experiments are conducted on two specific re-identification tasks (i.e., Person re-ID, Vehicle re-ID). Our ES-Net outperforms state-of-the-art methods on three Person re-ID benchmarks and two Vehicle re-ID benchmarks. Specifically, mAP / Rank-1 rate: 88.6% / 95.7% on Market1501, 78.8% / 89.2% on DuckMTMC-reID, 57.3% / 80.9% on MSMT17, 81.9% / 97.0% on Veri-776, respectively. Rank-1 / Rank-5 rate: 83.6% / 96.9% on VehicleID (Small), 79.9% / 93.5% on VehicleID (Medium), 76.9% / 90.7% on VehicleID (Large), respectively. Moreover, the visualized salient areas show human-interpretable visual explanations for the ranking results.
The width of a neural network matters since increasing the width will necessarily increase the model capacity. However, the performance of a network does not improve linearly with the width and soon gets saturated. To tackle this problem, we propose to increase the number of networks rather than purely scaling up the width. To prove it, one large network is divided into several small ones, and each of these small networks has a fraction of the original one's parameters. We then train these small networks together and make them see various views of the same data to learn different and complementary knowledge. During this co-training process, networks can also learn from each other. As a result, small networks can achieve better ensemble performance than the large one with few or no extra parameters or FLOPs. \emph{This reveals that the number of networks is a new dimension of effective model scaling, besides depth/width/resolution}. Small networks can also achieve faster inference speed than the large one by concurrent running on different devices. We validate the idea -- increasing the number of networks is a new dimension of effective model scaling -- with different network architectures on common benchmarks through extensive experiments. The code is available at \url{https://github.com/mzhaoshuai/SplitNet-Divide-and-Co-training}.