Capitalizing deep neural network with multifaceted semantic image segmentation integration methodology
The domain of unsupervised adaptation has always posed an intricate problem for the field of semantic image segmentation. The lack of information to predict each pixel has led to current research implementing the novelty of deep neural network applications to handle this issue. However, these method...
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/10356/162227 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Nanyang Technological University |
| 語言: | English |
| 總結: | The domain of unsupervised adaptation has always posed an intricate problem for the field of semantic image segmentation. The lack of information to predict each pixel has led to current research implementing the novelty of deep neural network applications to handle this issue. However, these methodologies are usually unimodal which, with proper integration strategies, form a deep multifaceted methodology that could achieve a better result. Thus, this paper has presented various unimodal along with conventional segmentation techniques that do not utilize the deep neural network. After which, the main methodology investigated possible integration techniques which encompassed early, late, and hybrid integration. A structured framework formulated from relevant datasets and performance benchmarks has been utilized to properly evaluate the results obtained. Limitations faced and a comprehensive evaluation of integration methodologies were discussed afterward to provide holistic insights as to when and how to utilize this integration methodology. |
|---|