三維列印又稱積層製造(Additive Manufacturing, AM)，可指任何列印三維物體的過程，又可分為七大製程，在市面上較為常見的方式為材料擠製成型(Material Extrusion, ME)。雖實驗室內已有將二維影像直接轉為三維列印所需G-code的新式建模[1]，但是在列印上與一般開源軟體一樣仍是以像素的灰階值來判斷並且分配固定的列印高度，在列印人像淺浮雕時會造成眼睛列印的比鼻子和嘴巴還要高，與實際人像的樣貌有所落差，本論文欲將深度學習與淺浮雕結合以解決上述問題。
本論文設計出一個人像淺浮雕軟體，應用CNN來訓練一個五官偵測的模型自動找尋人像中各個特徵(如：眼睛、鼻子與嘴巴) ，並將特徵座標輸出用於輔助人像淺浮雕軟體的分層高度設計中，使列印高度較為合理；記錄完每一層的座標點進行路徑規劃後直接生成G-code，跳過存成STL檔案的步驟可以避免STL檔案產生的缺點，提高列印品質，並且重新撰寫新式建模[1]，改善功能使其更為完整。

3D printing, also known as additive manufacturing (AM), can be expressed as any process of printing 3D objects. There are seven major methods for 3D printing. Fused deposition modeling (FDM) is the most common method of 3D printing. Although there is a new modeling that it could directly converts two-dimensional images into G-code for 3D printing in the laboratory, the printing is still determined by the gray scale value of the pixel to assign the fixed value for the printing height. When printing the portrait bas-relief, the height of the eyes will be higher than the nose and the mouth, it is different from the actual portrait.
This research design a portrait bas-relief software, which can find the characteristic in portraits automatically by using convolutional neural network (CNN) to train a feature detection modeling, and apply feature coordinate to help the design for layer height as well to make the printing height more reasonable. After recording the path planning of the coordinate on each layer, it will generate G-code directly, skipping the procedure of saving as STL files to avoid the disadvantages. In addition to the above, this results also regenerate the new modeling [1] to make the software function more complete.

[1] 曾采豐, “自動化二維影像建模與路徑規劃之研究,”國立成功大學航空太空工程學系碩士論文, 2019。
[2] 鄭正元,江卓培,林宗翰,林榮信,蘇威年,汪家昌,蔡明忠,賴維祥,鄭逸琳,洪基彬,鄭中緯,宋宜駿,陳怡文,賴信吉,吳貞興,許郁淞,陳宇恩, 3D列印基層製造技術與應用, 全華文化, 2017。
[3] https://www.opentech.tw/pages/documents/?/intro/C4-1.html
[4] 李昕叡, “三維列印與物聯網元件自動化整合研究, ” 國立成功大學航空太空工程學系碩士論文, 2018。
[5] J. Redmon, S. Divvala, R. Girshick and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, pp. 779-788, 2016.
[6] J. Hosang, R. Benenson, and B. Schiele, “Learning Non-maximum Suppression,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 6469-6477, 2017.
[7] Jean Hergel, Sylvain Lefebvre,“Clean Color: Improving Multi-filament 3D prints,” EUROGRAPHICS 2014, Vol. 33, no. 2, 2014.
[8] Y. LeCun, L. Bottu, Y. Bengio and P. Haffner, “Gradient-Based Learning Applied to Document Recognition,” in Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324,1998.
[9] A. Fred, M. Agarap, “Deep Learning Using Rectified Linear Units (ReLU),” arXiv:1803.08375, 2019.
[10] B. Xu, N. Wang, T. Chen, and M. Li, “Empirical Evaluation of Rectified Activations in Convolution Network,” arXiv preprint, arXiv: 1505.00853v2, [cs.LG] Nov. 2015.
[11] D. Scherer, A. Müller, and S. Behnke, “Evaluation of Pooling Operations in Convolutional Architectures for Object Recognition,” Proc. International conference on artificial neural networks, pp.92–101, 2010.
[12] R. Girshick, J. Donahue, T. Darrell, J. Malik. “Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation,” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 580–587, 2014.
[13] J. Redmon, & A. Farhadi, “YOLO9000: Better, Faster, Stronger, ” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pp. 7263-7271, 2017.
[14] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” In Advances in Neural Information Processing Systems, pp.91-99, 2015.
[15] Chigozie Nwankpa, Winifred Ijomah, Anthony Gachagan and Stephen Marshall, “Activation Functions: Comparison of Trends in Practice and Research for Deep Learning,”arXiv　preprint arXiv:1811.03378, 2018.
[16] J. Russakovsky , H. Deng, J. Su, S. Krause, S.Satheesh, Z. Ma, A. Huang, A. Karpathy, M. Khosla , A. Bernstein, C. Berg and L. Fei-Fei.“ImageNet Large Scale Visual Recognition Challenge,” International Journal of Computer Vision (IJCV), vol.115, no. 3, pp.211-252, 2015.
[17] I. Sutskever Krizhevsky, and G. E. Hinton, “Imagenet Classification with Deep Convolutional Neural Networks,” In Advances in neural information processing systems, pp.1097–1105, 2012.
[18] http://www.vision.caltech.edu/html-files/archive.html
[19] https://fei.edu.br/~cet/facedatabase.html
[20] Sebastian Ruder, “An Overview of Gradient Descent Optimization Algorithms,” arXiv preprint arXiv:1600.04747, 2017.
[21]https://zh.wikipedia.org/wiki/File:The_differences_between_CAD_and_STL_Models.svg
[22] 李享億, ”三維快速成型之切層技術研究,” 國立成功大學航太工程學系碩士論文,2003。
[23]http://www.paper.edu.cn/scholar/showpdf/MUT2AN1IMTT0Ix4h
[24] https://zh.wikipedia.org/wiki/%E5%83%8F%E7%B4%A0
[25] http://www.3up.com.tw/digitalimageSectionII.htm
[26] https://www.itread01.com/content/1545718930.html
[27] 陳榮昌, 盧菁, 劉賀平, “人臉座標系統的人臉識別,” 第十四屆人工智慧與應用研討會論文集, 台中，第30-31頁,2009。
[28] Bagus Arthaya , Ari Setiawan and Santoso Sunardi, “The Design and Development of G-Code Checker and Cutting Simulator for CNC Turning Operation,” Journal of Mechanical Engineering Research Vol. 2(3), pp. 58-70, 2010.
[29] http://infinity3dp.com/tw/html/product/show.php?pid=160&cid=38
[30] https://web.ncku.edu.tw/p/412-1000-49.php?Lang=zh-tw