隨著3D技術的快速發展，人們得以享有更真實的視覺體驗。在3D技術的許多應用當中，若希望以較低的成本達成目標，需要使用深度圖進行輔助，因為深度圖表示影像中每一像素與觀賞者的距離。在電腦視覺的領域中，單視角深度估計是非常熱門的議題，其優勢是只需要單張影像作為輸入即可進行深度估計。為了應用於2D轉3D的任務，我們希望能夠獲得相對深度圖，即物件之間的前後相對關係。另外，我們期望訓練模型不要有資料依賴的問題，能應用於多種不同場景。因此我們提出單視角深度估計網路進行深度估計。所提出的雙徑殘差網路鼓勵資訊共享，共同使用學習殘差圖與學習深度圖的兩路徑，目的是獲得更精準的深度圖。另外，我們使用了數個損失函數進行訓練，並針對各個損失函數進行加權，以獲得符合預期的深度圖。本篇論文的實驗結果顯示出我們所提出的方法不僅能在對訓練資料集進行測試時取得高品質的深度圖，在對其他場景進行深度估計時依舊有很出色的表現。

With the rapid development of 3D technology, people can enjoy a more realistic visual experience. In many applications of 3D technology, the use of depth maps, which represent the distance between each pixel in the image and the viewer, is necessary to achieve the goal at a lower cost. In the field of computer vision, monocular depth estimation is a very popular topic, with the advantage that only a single image is needed as input for depth estimation. For 2D to 3D tasks, we want to obtain relative depth maps, i.e., the anterior-posterior relationship between objects. In addition, we expect that the training model does not have data dependency problem and can be applied to many different scenes. We propose a monocular depth estimation network for depth estimation. The proposed dual-path residual network encourages information sharing by jointly using two paths of learning residual map and learning depth map, with the aim of obtaining a more accurate depth map. In addition, we use several loss functions for training and weight each loss function to obtain the expected depth map. The experimental results in this thesis show that our proposed method not only achieves high quality depth maps when testing the training dataset, but also performs well when estimating depths for other scenes.

[1] Zbontar, J., & LeCun, Y. (2015). Computing the stereo matching cost with a convolutional neural network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 1592-1599).
[2] Alatan, A. A., Yemez, Y., Gudukbay, U., Zabulis, X., Muller, K., Erdem, Ç. E., ... & Smolic, A. (2007). Scene representation technologies for 3DTV—A survey. IEEE Transactions on Circuits and Systems for Video Technology, 17(11), 1587-1605.
[3] Cho, H., Jung, S. U., & Jee, H. K. (2017, March). Real-time interactive AR system for broadcasting. In 2017 IEEE Virtual Reality (VR) (pp. 353-354). IEEE.
[4] Sharma, M., Chaudhury, S., & Lall, B. (2016). A novel hybrid kinect-variety-based high-quality multiview rendering scheme for glass-free 3D displays. IEEE Transactions on Circuits and Systems for Video Technology, 27(10), 2098-2117.
[5] Wang, P., Shen, X., Lin, Z., Cohen, S., Price, B., & Yuille, A. L. (2015). Towards unified depth and semantic prediction from a single image. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2800-2809).
[6] Laina, I., Rupprecht, C., Belagiannis, V., Tombari, F., & Navab, N. (2016, October). Deeper depth prediction with fully convolutional residual networks. In 2016 Fourth International Conference on 3D Vision (3DV) (pp. 239-248). IEEE.
[7] Fu, H., Gong, M., Wang, C., Batmanghelich, K., & Tao, D. (2018). Deep ordinal regression network for monocular depth estimation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2002-2011).
[8] Godard, C., Mac Aodha, O., & Brostow, G. J. (2017). Unsupervised monocular depth estimation with left-right consistency. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 270-279).
[9] Yin, Z., & Shi, J. (2018). Geonet: Unsupervised learning of dense depth, optical flow and camera pose. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 1983-1992).
[10] Zheng, C., Cham, T. J., & Cai, J. (2018). T2net: Synthetic-to-realistic translation for solving single-image depth estimation tasks. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 767-783).
[11] Ren, H., El-Khamy, M., & Lee, J. (2019, June). Deep Robust Single Image Depth Estimation Neural Network Using Scene Understanding. In Proceedings of CVPR Workshops (pp. 37-45).
[12] Kumari, S., Jha, R. R., Bhavsar, A., & Nigam, A. (2019, September). Autodepth: Single image depth map estimation via residual cnn encoder-decoder and stacked hourglass. In Proceedings of 2019 IEEE International Conference on Image Processing (ICIP) (pp. 340-344). IEEE.
[13] Alhashim, I., & Wonka, P. (2018). High quality monocular depth estimation via transfer learning. arXiv preprint arXiv:1812.11941.
[14] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 770-778).
[15] Ronneberger, O., Fischer, P., & Brox, T. (2015, October). U-net: Convolutional networks for biomedical image segmentation. In Proceedings of International Conference on Medical Image Computing and Computer-assisted Intervention (pp. 234-241). Springer, Cham.
[16] Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
[17] Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., ... & Rabinovich, A. (2015). Going deeper with convolutions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 1-9).
[18] Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. Q. (2017). Densely connected convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 4700-4708).
[19] Jeon, J., & Lee, S. (2018). Reconstruction-based pairwise depth dataset for depth image enhancement using CNN. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 422-438).