<form id="zdldf"></form><address id="zdldf"></address>

      <form id="zdldf"><nobr id="zdldf"></nobr></form>

        <address id="zdldf"></address>
        首頁 > 科學研究 > 學術看板 > 正文

        面向顯著性檢測的深度特征探索

        供稿:    責任編輯:安果    時間:2018-05-18    閱讀:

        主講人:張平平

        報告時間:201852111:00

        報告地點:基礎教學樓A303

        個人主頁:https://scholar.google.com/citations?user=MfbIbuEAAAAJ&hl=zh-CN

        報告題目:Delving into Deep Features for Saliency Detection

        (面向顯著性檢測的深度特征探索)


        報告摘要:

        顯著性檢測已經在計算機視覺的應用中取得了巨大的成功。然而,視覺顯著性的定義依賴于多種因素,很難用一種方式把所有的檢測線索有效地統一起來,因此準確的顯著性檢測仍是一個未解難題。本次報告將介紹我們近期在顯著性檢測的方面的工作,特別是利用深度網絡層次化的特點,探索不同深度全卷積網絡模型,進而實現顯著性檢測的一系列方法。其中包含:1)自適應地聚合多水平卷積特征進行復雜場景的顯著性物體檢測;2)學習深度不確定性卷積特征,提升物體邊界預測,進而提升物體檢測性能;3)利用金字塔池化獲取全局空間上下文特征,并進行逐步修正初始預測;4)根據圖像的本征反射,將輸入圖像進行適當的無損反射分解,提取互補特征,進行顯著物體檢測。我們的方法在公開的數據集上均取得了優于其他算法的性能。

        Saliency detection has achieved great success in computer vision applications. However, accurate saliency detection remains an unsolved problem because there are large variety of facts that cancontribute to define visual saliency, and it’s hard to combineall cues in an appropriate way. In this report, I will introduce our recent works on saliency detection, especially in different fully convolutional network models, which based on the hierarchical facts in deep neural networks. Our methods include:1) aggregating multi-level convolutionalfeature for salient object detection in complex scenes.2)learning deep uncertain convolutionalfeatures for boosting saliency detection, which encourage the confident boundaries of objects. 3) a stage-wise refinement model, in which a pyramid pooling moduleis applied for global context aggregation.4)based on the intrinsic reflectionof images, we decompose the input images into lossless reflection pairs to learn complementary features for saliency detection.Experimental evaluations on public benchmarksshow that our proposed methods compares favorably againstthe state-of-the-art approaches.


        報告人簡介:

        張平平博士現為澳大利亞視覺技術研究中心(ACVT)研究員. 他分別于2012年、2018年在河南師范大學、大連理工大學獲得理學學士、工學博士學位。師從大連理工大學盧湖川教授,其主要研究興趣為計算機視覺與機器學習。他已在國際計算機視覺和人工智能頂級會議(如ICCV,ECCV,IJCAI)以及期刊(如TIP,TCSVT,PR)上發表論文十數篇,并擔任多個會議及期刊的審稿人,如CVPR,ICCV,ECCV,IJCAI,TPAMI,IJCV,TIP等。

        Dr. Pingping Zhangis a research in Australian Centre of Visual Technology. In 2012 and 2018, He received the B.S. and Ph.D degree from Henan Normal University (HNU) and Dalian University of Technology (DUT), respectively. His supervisor is Prof. Huchuan Lu. His main research interests are in computer vision and machine learning. He has published more than 10 papers in top conferences/journals of computer vision and artificial intelligence, including ICCV, ECCV,IJCAI,TIP TCSVT,PR,etc. He also serves as the reviewer CVPR,ICCV,ECCVIJCAI,TPAMI,IJCV,TIP,etc..


        澳客时时彩官网