多重損失評估卷積網路於影像內容分割之應用

Multi-loss Convolutional Networks for Semantic Segmentation

林哲維(Che-Wei Lin)

指導教授 ： 陳煥宗

國立清華大學/電機資訊學院/資訊工程學系/碩士(2015年)

摘要

本論文研究的主題是場景物件辨識與切割。透過訓練深層卷積類神經網路,讓 照片中的每個像素都能夠被分類。與傳統非類神經網路為主的學習方法不同的是, 不需要取大量不同的特徵向量對於訓練類神經網路,我們提高了丟失率 (dropout rate) 以及利用不同的能量函數來衡量類神經網路的效能,比單一能量函數的衡量更 能提高對於小物件的辨識率。接著利用前述方法對於不同類別都有很高偵測率的特 性,設計了一套利用擴張小物體面積來讓小物體能更顯著被分類的方法。
最後藉由實驗結果探討改進之處,發現擴張小物體面積的技巧對於物體分割 的準確度的關係,以及了解多重能量函數對於類神經網路能否同時執行多項不同的 作業有無幫助。

關鍵字

卷積類神經網路 ； 影像內容分割 ； 多重損失評估

並列摘要

This thesis presents a semantic segmentation method based on fully-convolutional network (FCN). We focus on increasing mean-class accuracy by adding other steps that help FCN to find more small objects: i) modulating the dropout rates, ii) combining multiple loss functions, and iii) expanding small object areas.
Our approach shows that the above steps can significantly increase mean-class accuracy without sacrifice too much per-pixel accuracy. We also provide experimental observations on the relationship between the area-expanding method and the CNN model. Finally, we discuss how to improve the workflow and what we have learned from the experiments of training with multi-loss functions.

並列關鍵字

Convolutional Neural Network ； Semantic Segmentation ； Multi-loss

參考文獻

[1] Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille. Semantic image segmentation with deep convolutional nets and fully connected crfs. In ICLR, 2015.

[3] Mark Everingham, Luc J. Van Gool, Christopher K. I. Williams, John M. Winn, and Andrew Zisserman. The pascal visual object classes (VOC) challenge. International Journal of Computer Vision, 88(2):303–338, 2010.

[5] Cle ́ment Farabet, Camille Couprie, Laurent Najman, and Yann LeCun. Learning hierarchical features for scene labeling. IEEE Trans. Pattern Anal. Mach. Intell., 35(8):1915–1929, 2013.

[6] Pedro F. Felzenszwalb and Daniel P. Huttenlocher. Efficient graph-based image segmentation. International Journal of Computer Vision, 59(2):167–181, 2004.

[7] Marian George. Image parsing with a wide range of classes and scene-level context. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2015.