Digital makeup using machine learning algorithms
In this report, we present a pipeline system of digital makeup for industry scenarios. The pipeline contains two parts: i) facial feature semantic segmentation; ii) colour transfer. For facial feature semantic segmentation task, we adopt fully convolutional network (FCN) with weighted cross entropy...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | |
| 格式: | Final Year Project |
| 語言: | English |
| 出版: |
Nanyang Technological University
2022
|
| 主題: | |
| 在線閱讀: | https://hdl.handle.net/10356/156504 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | In this report, we present a pipeline system of digital makeup for industry scenarios. The pipeline contains two parts: i) facial feature semantic segmentation; ii) colour transfer. For facial feature semantic segmentation task, we adopt fully convolutional network (FCN) with weighted cross entropy as loss function during training; for colour transfer task, we experimented N-dimensional Probability Density Function transfer algorithm, a fast exemplar-based image colourisation approach using colour embeddings named Color2Embed, and deep exemplar-bases colourisation approach. Considering economical and qualitative factors, we conclude that model trained by VGG16 FCN with weighted cross entropy together with fast exemplar-based image colourisation yields the most suitable result. |
|---|