臺灣博碩士論文加值系統

室內設計產業在現今來說是不可或缺的一塊產業，許多時候客戶在裝潢住宅前會想預先看到自己房子內的真實樣貌，此時室內設計業就會提供擬真圖提供給客戶作為參考，擬真圖是利用建模軟體所繪製出來的。而建模的步驟為先設計出一個平面，接著拉高成為一個素模，再將素模繪製出有材質、有顏色、有燈光的3D模型，最後再利用渲染軟體將3D模型渲染為3D擬真圖。但從素模到3D擬真圖的繪製過程中會花費大量的時間以及視覺化思考。隨著時間的發展，人工智慧的技術也已經非常成熟，若擬真圖的設計可以利用人工智慧來製作將會省下許多時間以及視覺化思考等技術需求。本論文中使用生成對抗網路為基礎來學習室內設計的建模，將素模直接轉換為3D擬真圖，我們先利用SketchUp製圖蒐集大量的室內圖片，並使用Pix2pix以及CycleGAN對圖片進行訓練，視情況調整不同的訓練次數，從中分析出最佳結果接著與V-Ray渲染出來的圖片進行比較。實驗結果顯示，Pix2pix對於素模與擬真圖的轉換有較不錯的表現，相比之下CycleGAN並不適合對素模與擬真圖進行轉換。最後生成出來的最佳結果與V-Ray渲染出來的擬真圖進行比較後能夠發現，生成對抗網路模型產生的圖片的確能自動設計好房間，生成速度也非常快，整體的架構也非常清楚，但相比之下利用V-Ray渲染出來的圖片，生成對抗網路產生的圖片細節並沒有那麼明顯。最後我們希望透過本論文的實驗，能夠快速地將素模轉換為擬真圖，解決時間、技術、以及金錢問題，讓沒有設計能力的人也可以獲得所需要的圖片。

The interior design industry is an indispensable part of the industry today. Many times, customers want to see the real appearance of their houses before decorating their houses. At this time, the interior design industry will provide realistic pictures for customers as a For reference, the immersive map is drawn using modeling software. The modeling steps are to first design a plane, then pull it up to become a prime model, and then draw a 3D model with materials, colors and lights from the prime model, and finally use the rendering software to render the 3D model into a 3D simulation. real picture. However, it takes a lot of time and visual thinking in the process of drawing from the original model to the 3D realistic drawing. With the development of time, the technology of artificial intelligence has also become very mature. If the design of the simulation map can be made by artificial intelligence, it will save a lot of time and technical requirements such as visual thinking. In this paper, we use the generative adversarial network as the basis to learn the modeling of interior design, and directly convert the original model into a 3D realistic image. We first use SketchUp to collect a large number of indoor images, and use Pix2pix and CycleGAN to train the images. Adjust different training times according to the situation, analyze the best results and compare them with the pictures rendered by V-Ray. The experimental results show that Pix2pix has a good performance for the conversion of pixel simulation and simulation images. In contrast, CycleGAN is not suitable for the conversion of pixel simulation and simulation images. After comparing the best result generated with the realistic image rendered by V-Ray, it can be found that the image generated by the generative adversarial network model can indeed automatically design the room, the generation speed is also very fast, and the overall architecture is also very clear , but in contrast to the pictures rendered by V-Ray, the details of the pictures generated by the generative adversarial network are not so obvious. Finally, we hope that through the experiments of this paper, we can quickly convert the original model into a realistic image, solve the problems of time, technology, and money, so that people who have no design ability can also obtain the required images.

目錄
摘要 I
Abstract II
致謝 IV
目錄 V
圖目錄 VII
表目錄 IX
第一章 緒論 1
1.1 研究背景 1
1.2 研究動機與目的 2
第二章 文獻探討 6
2.1 室內設計三維模型 6
2.2 卷積神經網路(Convolutional Neural Network) 7
2.3 生成對抗網路(Generative Adversarial Network，GAN) 9
2.4 條件式生成對抗網路(conditional GAN，cGAN) 10
2.5 Pix2pix(圖像翻譯) 11
2.6 循環式生成對抗網路(CycleGAN) 14
第三章 研究方法 16
3.1 研究架構 16
3.2 資料蒐集 17
3.3 資料前處理 19
3.4 以Pix2Pix為基礎進行三維素模的轉換 20
3.5 以CycleGAN進行圖像風格轉換 23
3.6 參數設定 27
3.7 模型評估 28
第四章 實驗結果與分析 29
4.1 實驗環境 29
4.2 以Pix2pix為基礎的三維模型轉換結果 30
4.3 以 CylceGAN為基礎的三維模型轉換結果 42
4.4 實驗結果與訓練資料進行比較 48
第五章 結論與未來展望 52
5.1 結論 52
5.2 未來展望 55
參考文獻 57

圖目錄
圖1 1 室內設計素模 2
圖1 2 室內設計3D模型 3
圖1 3 室內設計3D擬真圖 3
圖1 4 三維立面圖像轉換為擬真的照片 4
圖2 1為SketchUp所建立的三維模型 7
圖2 2卷積層操作簡易架構 8
圖2 3為2x2平均池化層與最大值池化層 9
圖2 4自動編碼器簡易網路架構示意圖 10
圖2 5生成對抗網路的基本架構 10
圖2 6 cGAN基本架構 11
圖2 7 U-net基本架構 13
圖2 8 PatchGAN基本架構 13
圖2 9 Pix2Pix實際應用的範例圖 13
圖2 10 CycleGAN基本架構圖 15
圖2 11 CycleGAN應用範例 15
圖3 1研究流程圖 17
圖3 2素模與相同角度的擬真圖片 18
圖3 3水平翻轉過後的圖片 19
圖3 4將成對資料合併為256x512的新圖片 20
圖3 5 CycleGAN目標函式 23
圖4 1 實驗1生成結果細節處 34
圖4 2 實驗1生成結果產生雜訊處 34
圖4 3 epoch400次時牆壁紋路已成形 37
圖4 4 epoch400次時門的邊緣較為銳化 37
圖4 5 epoch400次時能夠生成出門把 38
圖4 6 第500次epoch生成出第三個門把 41
圖4 7 書桌epoch400次與epoch500次對比 41
圖4 8 臥室epoch400次與epoch500次對比 42


表目錄
表 1 編碼器詳細架構 21
表 2 解碼器詳細架構 21
表 3判別器詳細架構 22
表 4 生成器G編碼器詳細架構 24
表 5 生成器F編碼器詳細架構 24
表 6 生成器G解碼器詳細架構 24
表 7 生成器F解碼器架構 25
表 8 判別器詳細架構 26
表 9 參數設定 27
表 10 硬體設備及實驗環境 29
表 11 Pix2pix實驗數據 30
表 12 實驗1結果 31
表 13 epoch300次與epoch400次的比較 34
表 14 epoch400次與epoch500次的比較 38
表 15 CycleGAN實驗數據 42
表 16 實驗2結果 43
表 17 ecpoh300次與ecpoh400次的結果比較 46
表 18 Pix2pix生成結果與V-Ray渲染結果比較 49
表 19 對Pix2pix最佳結果進行分析 52

Antipov, G., Baccouche, M., & Dugelay, J. L. (2017, September). Face aging with conditional generative adversarial networks. In 2017 IEEE international conference on image processing (ICIP) (2089-2093). IEEE.
Bi, X., & Xing, J. (2020). Multi-Scale Weighted Fusion Attentive Generative Adversarial Network for Single Image De-Raining. IEEE Access, 8, 69838-69848.
Bu, Q., Luo, J., Ma, K., Feng, H., & Feng, J. (2020). An enhanced pix2pix dehazing network with guided filter layer. Applied Sciences, 10(17), 5898.
Chang, Y. L., Liu, Z. Y., Lee, K. Y., & Hsu, W. (2019). Free-form video inpainting with 3d gated convolution and temporal patchgan. In Proceedings of the IEEE/CVF International Conference on Computer Vision (9066-9075).
Demir, U., & Unal, G. (2018). Patch-based image inpainting with generative adversarial networks. arXiv preprint arXiv:1803.07422.
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., ... & Bengio, Y. (2014). Generative adversarial nets. Advances in neural information processing systems, 27.
He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask r-cnn. In Proceedings of the IEEE international conference on computer vision (2961-2969).
He, K., Sun, J., & Tang, X. (2012). Guided image filtering. IEEE transactions on pattern analysis and machine intelligence, 35(6), 1397-1409.
Isola, P., Zhu, J. Y., Zhou, T., & Efros, A. A. (2017). Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (1125-1134).
Karara, G., Hajji, R., & Poux, F. (2021). 3D Point Cloud Semantic Augmentation: Instance Segmentation of 360° Panoramas by Deep Learning Techniques. Remote Sensing, 13(18), 3647.
Kim, P. (2017). Convolutional neural network. In MATLAB deep learning (121-147). Apress, Berkeley, CA.
Li, G., Ma, B., He, S., Ren, X., & Liu, Q. (2020). Automatic tunnel crack detection based on u-net and a convolutional neural network with alternately updated clique. Sensors, 20(3), 717.
Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition ( 3431-3440).
Mirza, M., & Osindero, S. (2014). Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784.
Morgenstern, O., & Von Neumann, J. (1953). Theory of games and economic behavior. Princeton university press.
Ronneberger, O., Fischer, P., & Brox, T. (2015, October). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention ( 234-241). Springer, Cham.
Weng, W., & Zhu, X. (2015). Convolutional Networks for Biomedical Image Segmentation. IEEE Access.
Wu, H., Zhang, J., Huang, K., Liang, K., & Yu, Y. (2019). Fastfcn: Rethinking dilated convolution in the backbone for semantic segmentation. arXiv preprint arXiv:1903.11816.
Xie, C., Wang, Z., Chen, H., Ma, X., Xing, W., Zhao, L., ... & Lin, Z. (2021). Image Style Transfer Algorithm Based on Semantic Segmentation. IEEE Access, 9, 54518-54529.
Xu, Y., Wang, K., Yang, K., Sun, D., & Fu, J. (2019, September). Semantic segmentation of panoramic images using a synthetic dataset. In Artificial Intelligence and Machine Learning in Defense Applications (Vol. 11169, p. 111690B).
International Society for Optics and Photonics.
Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision ( 2223-2232).
Zhang, H., Xu, T., Li, H., Zhang, S., Wang, X., Huang, X., & Metaxas, D. N. (2017). Stackgan: Text to photo-realistic image synthesis with stacked generative adversarial networks. In Proceedings of the IEEE international conference on computer vision (5907-5915).
Zhao, Z. Q., Zheng, P., Xu, S. T., & Wu, X. (2019). Object detection with deep learning: A review. IEEE transactions on neural networks and learning systems, 30(11), 3212-3232.
于佩琴. (2014).。室內設計的本質性: 室內空間居家性之探討. 中原大學室內設計研究所學位論文, 1-83.
宋傑, 肖亮, 練智超, 蔡子贇, & 蔣國平. (2021).。基於深度學習的數字病理圖像分割綜述與展望. Journal of Software, 32(5).
冷翊 (2016)。以三維電腦繪圖為核心的室內設計流程及表現之研究。南華大學藝術與設計學院創意產品設計學系。
林庭生 (2021)。以Pix2Pix與超解析度成像網路為基礎之金門老照片修復研究。國立金門大學資訊科技與應用碩士班
施旻岳（2021）。以生成對抗網路為基礎之閩式建築風格轉換研究（碩士論文）。國立金門大學資訊科技與應用碩士班。
連禹睿 (2021)。 基於生成對抗網路 GAN 模型之書法字體生成系統.
張榮傑 (2015)。基於語義分割之影片風格轉換。國立交通大學多媒體工程研究所
張峻瑋 (2019).。3D 效果圖擬真度影響設計發展之視覺思考研究. 中原大學室內設計研究所學位論文, 1-184.
楊詒鈞 (2021)。生成對抗網路應用於多角度學習情緒辨識之研究(碩士論文)。國立中興大學資訊管理學系。
簡嘉琳 (2021)。基於生成對抗網路的繪畫風格轉換(碩士論文) 。國立宜蘭大學資訊工程學系研究所。