检索结果-内蒙古大学图书馆

Resolving Task Confusion in Dynamic Expansion Architectures for Class Incremental Learning

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Huang, Bingchen Chen, Zhineng Zhou, Peng Chen, Jiayin Wu, Zuxuan Shanghai Key Lab of Intelligent Information Processing School of Computer Science Fudan University China Shanghai Collaborative Innovation Center on Intelligent Visual Computing China University of Maryland College ParkMD United States

The dynamic expansion architecture is becoming popular in class incremental learning, mainly due to its advantages in alleviating catastrophic forgetting. However, task confusion is not well assessed within this framework, e.g., the discrepancy between classes of different tasks is not well learned (i.e., inter-task confusion, ITC), and certain priority is still given to the latest class batch (i.e., old-new confusion, ONC). We empirically validate the side effects of the two types of confusion. Meanwhile, a novel solution called Task Correlated Incremental Learning (TCIL) is proposed to encourage discriminative and fair feature utilization across tasks. TCIL performs a multi-level knowledge distillation to propagate knowledge learned from old tasks to the new one. It establishes information flow paths at both feature and logit levels, enabling the learning to be aware of old classes. Besides, attention mechanism and classifier re-scoring are applied to generate more fair classification scores. We conduct extensive experiments on CIFAR100 and ImageNet100 datasets. The results demonstrate that TCIL consistently achieves state-of-the-art accuracy. It mitigates both ITC and ONC, while showing advantages in battle with catastrophic forgetting even no rehearsal memory is reserved. 1 Copyright © 2022, The Authors. All rights reserved.

关键词： Distillation

Knowledge Graph Embedding Based Graph Convolutional Network for Link Prediction

学校读者我要写书评

暂无评论

Knowledge Graph Embedding Based Graph Convolutional Network ...

International Conference on Advanced Cloud and Big Data (CBD)

作者： Qiang Wu Qing He Henan Institute of Advanced Technology Zhengzhou University Zhengzhou P.R. China Key Lab of Intelligent Information Processing of Chinese Academy of Science (CAS) Institute of Computing Technology CAS Beijing China University of Chinese Academy of Science Beijing China

Knowledge Graph Embedding (KGE) aims to capture the inherent structural information within knowledge graphs (KGs) by means of representation learning. This is of paramount importance for various downstream tasks, including personalized recommendations, intelligent search, and relation extraction. Graph Convolutional Networks (GCNs), a type of neural network, have recently demonstrated significant success in modeling data structured as graphs. However, these models often overlook the intricate interrelation information between neighboring nodes, which can offer valuable insights for enhancing node embeddings. In this article, we present a knowledge graph embedding based graph convolutional network for link prediction. First, we utilize embeddings to represent entities and relations. Second, we perform circular correlation on subject embedding and relation embedding to generate two matrix and achieve full fusion of embedding at the element level. Third, we generate self-attention vectors using one of the matrices and perform the Hadamard product with the other matrix. Thereafter, new node features are generated through graph convolutional networks and residual unit. Finally, we use the scoring function to score the candidate triples. The experimental results demonstrate that our model achieves improved performance compared with a wide variety of baseline models.

关键词：

ObjectFormer for Image Manipulation Detection and Localization

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wang, Junke Wu, Zuxuan Chen, Jingjing Han, Xintong Shrivastava, Abhinav Lim, Ser-Nam Jiang, Yu-Gang Shanghai Key Lab of Intelligent Information Processing School of Computer Science Fudan University China Shanghai Collaborative Innovation Center on Intelligent Visual Computing China Huya Inc University of Maryland United States Meta AI

Recent advances in image editing techniques have posed serious challenges to the trustworthiness of multimedia data, which drives the research of image tampering detection. In this paper, we propose ObjectFormer to detect and localize image manipulations. To capture subtle manipulation traces that are no longer visible in the RGB domain, we extract high-frequency features of the images and combine them with RGB features as multimodal patch embeddings. Additionally, we use a set of learnable object prototypes as mid-level representations to model the object-level consistencies among different regions, which are further used to refine patch embeddings to capture the patch-level consistencies. We conduct extensive experiments on various datasets and the results verify the effectiveness of the proposed method, outperforming state-of-the-art tampering detection and localization methods. Copyright © 2022, The Authors. All rights reserved.

关键词： Embeddings

DFTR: Depth-supervised Fusion Transformer for Salient Object Detection

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Zhu, Heqin Sun, Xu Li, Yuexiang Ma, Kai Zhou, S. Kevin Zheng, Yefeng Key Lab of Intelligent Information Processing Chinese Academy of Sciences Institute of Computing Technology Beijing100190 China Tencent Jarvis Lab Shenzhen China University of Science and Technology of China Suzhou215123 China

Automated salient object detection (SOD) plays an increasingly crucial role in many computer vision applications. By reformulating the depth information as supervision rather than as input, depth-supervised convolutional neural networks (CNN) have achieved promising results on both RGB and RGB-D SOD scenarios with the merits of no requirements for extra depth networks and depth inputs in the inference stage. This paper, for the first time, seeks to expand the applicability of depth supervision to the Transformer architecture. Specifically, we develop a Depth-supervised Fusion TRansformer (DFTR), to further improve the accuracy of both RGB and RGB-D SOD. The proposed DFTR involves three primary features: 1) DFTR, to the best of our knowledge, is the first pure Transformer-based model for depth-supervised SOD;2) A multi-scale feature aggregation (MFA) module is proposed to fully exploit the multi-scale features encoded by the Swin Transformer in a coarse-to-fine manner;3) To enable bidirectional information flow across different streams of features, a novel multi-stage feature fusion (MFF) module is further integrated into our DFTR with the emphasis on salient regions at different network learning stages. We extensively evaluate the proposed DFTR on ten benchmarking datasets. Experimental results show that our DFTR consistently outperforms the existing state-of-the-art methods for both RGB and RGB-D SOD tasks. The code and model will be made publicly available. © 2022, CC BY-NC-SA.

关键词： Object detection

Automatic Construction of Classification Dimensions on Texts

学校读者我要写书评

暂无评论

SSRN

SSRN 2022年

作者： Ma, Bing Zhuge, Hai The Key Lab of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences Beijing China University of Chinese Academy of Sciences Beijing China The Great Bay University Dongguan China

A way to efficiently manage large sets of texts is to construct multiple dimensions on texts and manage the texts from different dimensions. An approach to automatically constructing classification dimensions is to generate class trees on texts and a label for each class so that texts can be efficiently managed based on the generated trees and labels. The key issue is to classify texts into classes according to common semantics defined by a representation model of text. The vector space model represents texts in a space of high dimensions as it regards word as dimension. The classification methods based on the vector space is to classify texts according to the calculation of the distance between vectors of texts. However, the characteristics of the distances like the triangle inequality of distances is unsuitable to restrict the semantic relations between texts and measure the similarity of a set of more than two texts. The frequently occurred common words within a set of texts is a way to represent the texts but the traditional frequent-term-set based text clustering approaches generate a wide range of frequent term sets from one word to a maximum set of words satisfying the pre-determined minimum support degree, the initial classes it generates are not in line with the original classes, and it often generates the meaningless initial classes according to frequent-term sets. Topic model is not suitable for generating class trees on texts as topics are represented by probability distribution on all words. This paper proposes an approach to use common words of texts to represent the texts and calculates the similarity of a class of texts by calculating the sum of the weights of common words of the class where the common words of a class of texts represent the common semantics of these texts and the idea is in line with human classification of texts. Based on the common words representation, a bottom-up text clustering approach is proposed to construct class trees of t

关键词： information management

DRAG: Dynamic Region-Aware GCN for Privacy-Leaking Image Detection

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Yang, Guang Cao, Juan Sheng, Qiang Qi, Peng Li, Xirong Li, Jintao Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China Key Lab of Data Engineering and Knowledge Engineering Renmin University of China China

The daily practice of sharing images on social media raises a severe issue about privacy leakage. To address the issue, privacy-leaking image detection is studied recently, with the goal to automatically identify images that may leak privacy. Recent advance on this task benefits from focusing on crucial objects via pretrained object detectors and modeling their correlation. However, these methods have two limitations: 1) they neglect other important elements like scenes, textures, and objects beyond the capacity of pretrained object detectors;2) the correlation among objects is fixed, but a fixed correlation is not appropriate for all the images. To overcome the limitations, we propose the Dynamic Region-Aware Graph Convolutional Network (DRAG) that dynamically finds out crucial regions including objects and other important elements, and models their correlation adaptively for each input image. To find out crucial regions, we cluster spatially-correlated feature channels into several region-aware feature maps. Further, we dynamically model the correlation with the self-attention mechanism and explore the interaction among the regions with a graph convolutional network. The DRAG achieved an accuracy of 87% on the largest dataset for privacy-leaking image detection, which is 10 percentage points higher than the state of the art. The further case study demonstrates that it found out crucial regions containing not only objects but other important elements like textures. The code and more details are in https://***/guang-yanng/DRAG. © 2022, CC BY.

关键词： Object detection

Efficient Facial Landmark Detector by Knowledge Distillation

学校读者我要写书评

暂无评论

Efficient Facial Landmark Detector by Knowledge Distillation

International Conference on Automatic Face and Gesture Recognition

作者： Yuyang Sha Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS) Institute of Computing Technology Beijing CAS China

ISBN: (纸本)9781665431774

Facial landmark detection aims to locate some predefined points on human face images, which is the basis of many facial analysis tasks and applications. Compared with traditional methods, facial landmark detector based on deep learning has made significant progress in accuracy and efficiency. However, the model parameters and computational cost have not received enough attention, leading to the algorithm facing many challenges in practical applications. In this paper, we propose a novel method named efficient facial landmark detector (EFLD), which significantly compresses the model size and computational cost while achieving superior performance compared with state-of-the-art frameworks. We introduce a knowledge distillation framework suitable for facial landmark detection, which can transfer geometric information from the pre-trained teacher model to the lightweight student model. In addition, we integrate adversarial learning into the knowledge distillation framework to guide the student model to imitate the heatmap distribution generated by the teacher model. Extensive experiments demonstrate the superior performance of our approach against state-of-the-art methods.

关键词： Knowledge engineering Heating systems Computational modeling Face recognition Neural networks Detectors Gesture recognition

Towards Occlusion Robust Facial Landmark Detector

学校读者我要写书评

暂无评论

Towards Occlusion Robust Facial Landmark Detector

International Conference on Automatic Face and Gesture Recognition

作者： Yuyang Sha Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS) Institute of Computing Technology CAS Beijing China

ISBN: (纸本)9781665431774

Facial landmark detection plays an essential role in many face analysis tasks, such as face recognition, facial action unit detection, etc. However, when occlusion occurs, the existing methods might not perform well. This paper studies a new framework for facial landmark detection referred to as the Occlusion Robust Facial Landmark Detector (OR-FLD), which attempts to solve the occlusion problem. Specifically, the proposed OR-FLD consists of three main modules: Feature Random Dropping Module (FRD), Feature Activation Diversity Module (FAD), and Landmark Refinement Attention Module (LRA). The FRD forces the model to focus on the geometric relationships between different facial components by randomly dropping some appearance information and then automatically recovering them. The FAD tries to enhance the model representation and make the extracted features discriminative and robust to occlusions. Besides, the recovered features contribute differently to the final results. Therefore, we propose a simple and effective attention mechanism to refine the landmark location and make the model more robust to occlusions. We extensively evaluate OR-FLD on four facial landmark detection benchmarks (300W, WFLW, COFW, and AFLW). The results show that our method outperforms the state-of-the-art by a significant margin, especially on the datasets with large occlusion.

关键词： Location awareness Sensitivity Face recognition Pose estimation Detectors Gesture recognition Benchmark testing

Pattern Synthesis of MIMO Radar Based on Immune Differential Evolution Algorithm

学校读者我要写书评

暂无评论

Pattern Synthesis of MIMO Radar Based on Immune Differential...

2020 International Conference on Modeling, Big Data Analytics and Simulation, MBDAS 2020

作者： Jiang, Ling Jiang, Yi Chen, Guangfu Shi, Guimin Cheng, Qi School of Mathematics and Computer Wuyi University Wuyishan354300 China Key Lab. of Cogn. Computing and Intelligent Information Processing of Fujian Education Institutions Wuyishan354300 China Big Data Development Administration of Anji County Huzhou313000 China

In order to optimize the pattern synthesis of multiple input and multiple output (MIMO) radar, immune mechanism is adopted to overcome the premature risk of differential evolution (DE) algorithm, namely immune differential evolution (IDE). Firstly, the modeling of MIMO radar is introduced by encoding the position of array in the binary. Secondly, the immune mechanism is employed to improve the DE. In IDE, two parameters are self-adapted for the mutation operator by immune mechanism to enhance the convergence ability of DE, including scaling factor and crossover rate. Several experiments are conducted to analysis the performance of IDE. The simulation results show that the IDE with variable parameters can get optimal results in MIMO radar with lower the peak side-lobe level (PSLL) and maintain the diversity with stronger convergence ability and shorter calculation speed. © Published under licence by IOP Publishing Ltd.

关键词： Optimization