臺灣博碩士論文加值系統

房間空間佈局預測是輸入一張房間圖片而輸出對應面標籤。之前的方法是預測一組房間的角落點和其對應的房間類型，然後再根據房間類型將事先定義好的坊間角落順序依序連線產生其平面標籤圖。這個方法要求要事先定義房間的類型。 在本篇論文中，我們提出一個基於平面切割之全卷積神經網路自動編碼器來預測房間空間佈局。以一張圖片當作輸入資料，直接輸出一個平面切割的端到端可訓練自動編碼器。我們使用了省略與連接的技巧增強了輸出結果的準確率，還使用了多目標訓練方式訓練了平面分割及平面邊線熱度圖，平面邊線熱度圖的目的是為了學習到直線的特徵和細節。相較於之前的方法，我們的方法並不需要事先定義房間類型，只需要標簽好對應平面的資料集，而且可以處理擁有非直線牆角的房間架構類型。

Room layout estimation from the monocular RGB image is the purpose of the paper. The state-of-the-art method first estimates an ordered set of room layout key points and room types and then generates the corresponding segmentation image of room layout based on them. However, the method requires predefining 11 room types which are actually a limitation. Therefore, we want to directly estimate the segmentation image of room layout without doing the predefinition and would not be restricted to room types. In this paper, an end-to-end trainable Autoencoder is built to achieve the goal. Moreover, we adopt two ideas while building Autoencoder. One is skip connected architecture called modified U-net, and the other is multi-task which aims to generate both the boundaries contour map and the segmentation image. The experiment results show that ours can lead to better room layout estimations which are not limited to the room types.

誌謝 i
中文摘要 ii
ABSTRACT iii
CONTENTS iv
LIST OF FIGURES vi
LIST OF TABLES viii
Chapter 1 Introduction 1
1.1 Introduction of room layout estimation 1
1.2 Motivation 4
Chapter 2 Related Work 6
Chapter 3 Method 10
3.1 Choosing room layout representation 10
3.2 Autoencoder 11
3.3 Modified U-net 13
3.4 Multi-task 14
Chapter 4 Experimental Results 16
4.1 Datasets 16
4.2 Implementation details 16
4.3 Results 17
4.4 Runtime 19
4.5 Analyzing 19
4.6 More results 24
4.7 Cross Validation 29
4.8 Discussion 29
Chapter 5 Experimental Results 31
5.1 Conclusions 31
5.2 Future work 31
REFERENCE 32

[1]V. Hedau, D. Hoiem, and D. Forsyth. Recovering the Spatial Layout of Cluttered Rooms. In International Conference on Computer Vision (ICCV), 2009.
[2]D. Hoiem, A. A. Efros, and M. Hebert. Recovering surface layout from an image. In International Journal of Computer Vision (IJCV), 2007.
[3]D. C. Lee, M. Hebert and T. Kanade. Geometric Reasoning for Single Image Structure Recovery. In Computer Vision and Pattern Recognition (CVPR), 2009.
[4]A. Mallya and S. Lazebnik. Learning Informative Edge Maps for Indoor Scene Layout Prediction. In International Conference on Computer Vision (ICCV), 2015.
[5]S. Dasgupta, K. Fang, K. Chen, and S. Savarese. Delay: Robust Spatial Layout Estimation for Cluttered Indoor Scenes. In Computer Vision and Pattern Recognition (CVPR), 2016.
[6]M. E. Tipping, C. M. Bishop, Baysian image super-resolution, Neural Information Processing Systems (NIPS), 2003.
[7]C. Y. Lee, V. Badrinarayanan. T. Malisiewicz and A. Rabinovich. RoomNet: End-to-End Room Layout Estimation. arXiv:1703.06241v1 2017
[8]Y. Zhang, F. Yu, S. Song, P. Xu, A. Seff, and J. Xiao. Largescale Scene Understanding Challenge: Room layout estimation, accessed on Sep 15 (2015).


[9]O. Ronneberger, P. Fischer, and T. Brox. U-net: Convolutional Networks for Biomedical Image Segmentation. In Medical Image Computing and Computer Assisted Intervention (MICCAI), Springer, 2015.
[10]V. B adrinarayanan, A. Kendall, and R. Cipolla. "Segnet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation.” arXiv: 1511.00561 2016
[11]Y. Ren, C. Chen, S. Li, and C.-C. J. Kuo. A Coarse-to-Fine Indoor Layout Estimation (CFILE) Method. In Asian Conference on Computer Vision (ACCV) 2016