Cross-view scene image localization with Triplet Network integrating NetVLAD and Fully Connected Layers

نویسندگان

چکیده

ç ”ç©¶åœºæ™¯å›¾åƒçš„åœ°ç†å®šä½é—®é¢˜åœ¨å®¤å¤–å®šä½ã€ç›®æ ‡æœå¯»ã€å†›äº‹ä¾¦å¯Ÿç­‰é¢†åŸŸå ·æœ‰é‡è¦æ„ä¹‰ã€‚é’ˆå¯¹è¡—æ™¯å½±åƒä¸Žé¸Ÿçž°å½±åƒä¹‹é—´çš„äº¤å‰è§†è§’åœºæ™¯å›¾åƒåŒ¹é ä¸Žå®šä½é—®é¢˜ï¼Œæœ¬æ–‡æå‡ºäº†ä¸€ç§èžåˆå¯è®­ç»ƒå±€éƒ¨èšé›†æè¿°å­å‘é‡NetVLAD(Net Vector of locally aggregated descriptorsï¼‰å’Œå ¨è¿žæŽ¥å±‚çš„ä¸‰å ƒç¥žç»ç½‘ç»œï¼ˆTriplet Network)定位方法(Tri-NetVLADï¼‰ã€‚ä¸‰å ƒç¥žç»ç½‘ç»œç”±ä¸‰ç»„å·ç§¯ç¥žç»ç½‘ç»œCNN(Convolutional Neural Networks)构成,能同时处理3å¼ å½±åƒï¼Œé€šè¿‡å¢žå¤§ä¸åŒ¹é åƒå¯¹é—´çš„è·ç¦»ï¼Œå‡å°åŒ¹é åƒå¯¹é—´çš„è·ç¦»ï¼Œå®žçŽ°å›¾åƒæ£€ç´¢ä¸ŽåŒ¹é ï¼›NetVLADå’Œå ¨è¿žæŽ¥å±‚çš„èžåˆå¯ä»¥åŠ å¼ºç‰¹å¾é—´çš„å ³è”æ€§ã€‚æœ¬æ–‡å°†CNN提取的局部卷积特征分别通过NetVLADå±‚å’Œå ¨è¿žæŽ¥å±‚å¾—åˆ°å ¨å±€æè¿°ç¬¦ä¸Žç‰¹å¾å‘é‡ï¼Œå¹¶å°†äºŒè€ èžåˆï¼Œæœ‰æ•ˆåœ°æå‡äº†å±€éƒ¨ç‰¹å¾é—´çš„å ³è”æ€§ï¼Œå¹¶ä¿ç•™äº†ä¸åŒå±€éƒ¨ç‰¹å¾ä¹‹é—´çš„å·®å¼‚æ€§ï¼Œæå‡äº†æ¨¡åž‹çš„å®šä½ç²¾åº¦ï¼›æ”¹è¿›äº†DBL loss(Distance-based layer lossï¼‰ï¼Œé€šè¿‡åŠ å ¥å‚æ•°Î»å¢žå¼ºå‡½æ•°åˆ¤åˆ«å›°éš¾æ ·æœ¬çš„èƒ½åŠ›ï¼Œåœ¨æå‡æ¨¡åž‹çš„æ”¶æ•›é€Ÿåº¦å’Œç¨³å®šæ€§çš„åŒæ—¶ä¹Ÿæå‡äº†æ¨¡åž‹çš„å®šä½ç²¾åº¦ã€‚åœ¨ç¾Žå›½Vo and Hayså ¬å¼€æ•°æ®é›†ä¸Šçš„å®žéªŒç»“æžœè¡¨æ˜Žï¼ŒTri-NetVLAD取得了优于MCVPlaces、Triplet eDBL-Net和CVM-Net等现有方法的定位精度,在测试集上的精度高于63%。

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Estimation of Network Reliability for a Fully Connected Network with Unreliable Nodes and Unreliable Edges using Neuro Optimization

In this paper it is tried to estimate the reliability of a fully connected network of some unreliable nodes and unreliable connections (edges) between them. The proliferation of electronic messaging has been witnessed during the last few years. The acute problem of node failure and connection failure is frequently encountered in communication through various types of networks. We know that a ne...

متن کامل

Transitioning Between Convolutional and Fully Connected Layers in Neural Networks

Digital pathology has advanced substantially over the last decade however tumor localization continues to be a challenging problem due to highly complex patterns and textures in the underlying tissue bed. The use of convolutional neural networks (CNNs) to analyze such complex images has been well adopted in digital pathology. However in recent years, the architecture of CNNs have altered with t...

متن کامل

Scalable Scene Reconstruction and Image Based Localization

In this thesis two fundamental problems in computer vision are addressed: robust and scalable structure from motion and efficient localization from images. These two problems are highly interrelated tasks with several industrial applications, like mapping, navigation and augmented reality. The main contribution of this thesis is in building a complete, robust and scalable image based reconstruc...

متن کامل

In Defense of Fully Connected Layers in Visual Representation Transfer

Pre-trained convolutional neural network (CNN) models have been widely applied in many computer vision tasks, especially in transfer learning tasks. In transfer learning, the target domain may be in a different feature space or follow a different data distribution, compared to the source domain. In CNN transfer tasks, we often transfer visual representations from a source domain (e.g., ImageNet...

متن کامل

Deep Neural Networks In Fully Connected CRF For Image Labeling With Social Network Metadata

We propose a novel method for predicting image labels by fusing image content descriptors with the social media context of each image. An image uploaded to a social media site such as Flickr often has meaningful, associated information, such as comments and other images the user has uploaded, that is complementary to pixel content and helpful in predicting labels. Prediction challenges such as ...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: Journal of remote sensing

سال: 2021

ISSN: ['1007-4619', '2095-9494']

DOI: https://doi.org/10.11834/jrs.20210188