- Jobs
ResumeCreate your job-winning resume using our free resume builder.
PortfolioShowcase your skills and projects with a professional portfolio.ResumeCreate your job-winning resume using our free resume builder.Resume BuilderMake a resume for free.Resume TemplatesAccess our extensive library of professional & ready-to-use templates.Resume ExamplesGet inspired by real resume examples to create your own.Occupation GuideAccess resume writing guides tailored for different professions.Resume HelpGet expert advice on all things resume from our team of recruitment specialists.- ResourcesSuccess StoriesBusiness ExcellenceAbout CakeResumeFeatured Reads
- Hire
- Download our App
太空影像模組深度學習應用
太空影像模組深度學習應用
Senior Software Engineer
・
Taipei City, Taiwan
此專案針對曝光度不足之太空影像擷取模組所開發所開發之深度學習影像增強演算法
參考https://li-chongyi.github.io/Proj_Zero-DCE.html
基於硬體限制修改網路深度、損失函數與Short cut架構。主要將此演算法應用於ARM Cortex M3，組合出來的浮點數運算效率差，因此須將全網路運算量化。
全網路量化包含初期標準化、卷積運算、Sigmoid區段線性映射、雙線性內插以及最後的深度曲線迭代。
使用深度學習補足衛星影像因曝光短而亮度不足之問題
- 使用Zero Reference影像增強網路模型
- 使用圖像增強前後差異或目標值做為損式函數
- 不對影像資料標籤
- 配合硬體限制減少模型深度與網路尺度
- 將神經網路底層運算建立在ARM Cortex M3程式
- 實現量化整數運算(記憶體用量與運算量減少)
- 影像解析度 1920 x 1200
- Floating : 18 min per frame
- INT : 1 min per frame
Please login to comment.
