| DC 欄位 |
值 |
語言 |
| DC.contributor | 通訊工程學系 | zh_TW |
| DC.creator | 郭祐昇 | zh_TW |
| DC.creator | You-Sheng Guo | en_US |
| dc.date.accessioned | 2022-8-4T07:39:07Z | |
| dc.date.available | 2022-8-4T07:39:07Z | |
| dc.date.issued | 2022 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=109523046 | |
| dc.contributor.department | 通訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 近年人工智慧的發展迅速，各個產業紛紛透過機器取代或輔助人力，降低生產成本。在遊戲的領域中， 為了使人物的自然度更貼近現實生活，遊戲開發者需與動畫設計師共同研發 三維 人體模型 ，但是花費的時間與金錢過 高，提高開發成本，於是以 深度學習 去研發 三維 人體模型而不需要掃瞄儀器的輔助可以大幅降低遊戲開發成本。本研究將單張影像重建三維人體模型並以深度學習方式進行訓練 ，且在少量的資料集中達到高品質的重建。 近期文獻都以大量的資料進行訓練，不僅花費大量時間與提高購買訓練資料的成本，且無法供應個人使用。為了配合少量資料集進行模型訓練，本研究調整網路架構，使其能適應低資料庫訓練 ，可以確保非公司企業之個人使用該 三維 人體模型。 模型加入注意力機制使其在訓練時提取重要的特徵 提高重建三維人體模型的品質以及減少參數更新的時間， 另外，重建的模型不單只有幾何（Geometry）而是有顏色上的表現，能應用更廣泛。 本研究 不管是在客觀的評估（Point to Surface、 Chamfer Distance）或者重建 三維 人體模型的評估，兩者都有傑出的表現。
關鍵字: 重建三維人體模型、注意力機制、深度學習 | zh_TW |
| dc.description.abstract | With the rapid development of artificial intelligence in recent years, various industries have been replacing or aiding manpower through machines to reduce production costs. In order to make the naturalness of the characters closer to the real life, game developers need to develop 3D human models together with animation designers, but the time and money spent are too high, which increases the development cost. Therefore, using deep learning to develop 3D human models without the assistance of scanning instruments can significantly reduce game development costs. In the research, the 3D human model is reconstructed from a single image and trained with deep learning to achieve a high quality reconstruction with a small dataset. Recent literature has trained with a large amount of data, which not only takes a lot of time and increases the cost of purchasing training materials, but is also not available for personal use. In order to train the model with a small number of datasets, this study adapted the network architecture to accommodate low database training, which can ensure the use of the 3D human model by individuals in non- corporate enterprises. The addition of Attention to the model allows it to extract important features during training, improving the quality of the reconstructed 3D human model and reducing the time it takes to update parameters. In addition, the reconstructed model has not only geometry but also color representation, which can be used in a wider range of applications. Both have outstanding performance in objective evaluation or evaluation of reconstructed 3D human models. | en_US |
| DC.subject | 重建 三維 人體模型 | zh_TW |
| DC.subject | 注意力機制 | zh_TW |
| DC.subject | 深度學習 | zh_TW |
| DC.subject | Reconstruction of 3D Human Body Model | en_US |
| DC.subject | Attention Mechanism | en_US |
| DC.subject | Deep Learning | en_US |
| DC.title | 具有注意力機制之隱式表示於影像重建 三維人體模型 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Implicit Representation with Attention Mechanism for Image Reconstruction of 3D Human Model | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |