LabelImg dataset的問題，透過圖書和論文來找解法和答案更準確安心。 我們找到下列免費下載的地點或者是各式教學

深度學習影像資料集剪裁及標記方法及其於台灣交通號誌辨識之應用

為了解決LabelImg dataset的問題，作者林柏翰 這樣論述：

為了提供深度學習架構訓練模型，可能需要與取自於不同管道的影像資料集不同長寬比和解析度的圖片，此時需要花費大量人力和時間，對原始圖片進行剪裁，再對新圖片中的物件進行標記。我們提出一個用於深度學習影像資料集的剪裁和標記方法。針對不同長寬比和解析度圖像的需求，我們所提出的方法能夠計算剪裁範圍所能涵蓋最多標記的最佳標記組合與最佳剪裁區域，因此具有快速剪裁和重新標記影像中物件的能力，並可得到最多的訓練資訊。本論文將以我們所提出的剪裁及標記方法分別應用在比利時交通號誌資料集與我們自己建立的台灣交通號誌資料集的號誌辨識。經由統計發現我們的方法與固定剪裁範圍的對照組相比，我們提出的方法確實能夠在有限的剪裁範

圍保留最多的影像與標記，保留下的訓練影像及標記分別可達到99.9%與90.5%。利用剪裁後的影像資料集經由YOLOv3訓練後並偵測測試資料集，我們的方法相較於對照組在mAP上有24.8%至32.8%的提升；此外，在相同數量的影像資料集中，我們的方法在精確率上也有20.1%至23.3%的提升。以此證明我們的方法所剪裁及標記的影像資料集，對於深度學習架構在訓練模型時，具有訓練品質較好、mAP較佳的優勢。而且對於人力成本，我們的方法能節省許多時間與人工，平均處理一張耗費約0.85秒，相較於人工所花費的時間成本耗費節省了約47倍時間。

使用深度學習方法分析ADPKD患者的非顯影和顯影電腦斷層圖像的電腦視覺任務

為了解決LabelImg dataset的問題，作者Djeane Debora Onthoni 這樣論述：

ContentsABSTRACT. . . . . . . iTABLE OF CONTENTS. . . . . . . iiLIST OF FIGURES. . . . . . . viLIST OF TABLES. . . . . . . viiiLIST OF ABBREVIATIONS. . . . . . . ix1 Introduction 11.1 Medical Imaging . . . . . . . . . . . . . . . . . . 11.1.1 Ultrasound . . . . . . . . . . . . . . . . . . . 11.1.2

Magnetic Resonance Imaging . . . . . . . . . . . 21.1.3 Computed Tomography . . . . . . . . . . . . . . . 31.2 Artificial Intelligence . . . . . . . . . . . . . . 51.2.1 Machine Learning . . . . . . . . . . . . . . . . 51.2.1.1 Supervised Learning . . . . . . . . . . . . . . 51.2.1.2 Unsupervised

Learning . . . . . . . . . . . . . 61.2.2 Deep Learning . . . . . . . . . . . . . . . . . . 71.2.2.1 Classification Task . . . . . . . . . . . . . . 81.2.2.2 Localization Task . . . . . . . . . . . . . . . 91.2.2.3 Segmentation Task . . . . . . . . . . . . . . . 101.3 Kidney Disease . . . . . . . .

. . . . . . . . . . 111.4 Motivations . . . . . . . . . . . . . . . . . . . . 131.4.1 Non-contrast-enhanced Computed Tomography . . . . 141.4.2 Contrast-enhanced Computed Tomography . . . . . . 151.4.3 Localization and Segmentation for analyzing TKV . 151.5 Main Contributions . . . . . . . . . . .

. . . . . 161.6 Thesis Organization . . . . . . . . . . . . . . . . 162 Related works 182.1 Without Artificial Intelligence . . . . . . . . . . 182.2 With Artificial Intelligence . . . . . . . . . . . 192.2.1 Localization of ADPKD . . . . . . . . . . . . . . 192.2.2 Segmentation of ADPKD . . . .

. . . . . . . . . . 213 Automatic ADPKD Kidneys Localization Model on NCCT and CCT 233.1 Introduction . . . . . . . . . . . . . . . . . . . 233.2 Materials and Methods . . . . . . . . . . . . . . . 243.2.1 Data Acquisition . . . . . . . . . . . . . . . . 243.2.2 Ground Truth Annotation . . . . . .

. . . . . . . 253.2.3 Methods . . . . . . . . . . . . . . . . . . . . . 253.2.3.1 Preprocessing . . . . . . . . . . . . . . . . . 253.2.3.2 Dataset Partition . . . . . . . . . . . . . . . 273.2.3.3 Bounding Box Labeling . . . . . . . . . . . . . 283.2.3.4 Automatic ADPKD Kidneys Localization Model

. . .283.2.3.5 Training and Tuning Model . . . . . . . . . . . 303.2.3.6 Image-Wise and Subject-Wise Testing and Evaluation . . 313.2.4 Experimental Setup . . . . . . . . . . . . . . .. 313.2.5 Evaluation Metrics . . . . . . . . . . . . . . . 313.2.6 Evaluation Procedures . . . . . . . . . . . . .

. 323.3 Results on NCCT and CCT . . . . . . . . . . . . . . 333.3.1 Evaluation Results of Validation set on NCCT . . 333.3.2 Evaluation Results of Testing set on NCCT . . . . 333.3.3 Evaluation Results of Validation Set on CCT . . . 343.3.4 Evaluation Results of Testing set on CCT . . . . 353.4 Ev

aluation Results of Image-Wise Testing . . . . . 363.5 Evaluation Results of Subject-Wise Testing . . . . 383.6 Discussion . . . . . . . . . . . . . . . . . . . . 403.7 Conclusion . . . . . . . . . . . . . . . . . . . . 464 Automatic ADPKD kidneys Segmentation Model and TKV Estimation Modelon NC

CT and CCT 484.1 Introduction . . . . . . . . . . . . . . . . . . . 484.2 The Proposed Method . . . . . . . . . . . . . . . . 504.2.1 Data Preprocessing . . . . . . . . . . . . . . . 504.2.2 Automatic ADPKD Kidneys Segmentation . . . . . . 514.2.3 TKV Estimation Model . . . . . . . . . . . . . .

534.3 Experiment and Results . . . . . . . . . . . . . . 544.3.1 Dataset . . . . . . . . . . . . . . . . . . . . . 544.3.2 Experimental Setup . . . . . . . . . . . . . . . 554.3.3 Evaluation Metrics . . . . . . . . . . . . . . . 564.3.4 ADPKD Kidney Segmentation Results . . . . . . . . 564.3.4.1

Validation set results on NCCT . . . . . . . . 574.3.4.2 Testing set results on NCCT . . . . . . . . . . 57vii4.3.4.3 Validation set results on CCT . . . . . . . 584.3.4.4 Testing set results on CCT . . . . . . . . . . 594.3.5 TKV Estimation Results . . . . . . . . . . . . . 604.4 Discussion .

. . . . . . . . . . . . . . . . . . . 604.5 Conclusion . . . . . . . . . . . . . . . . . . . . 645 Conclusions and Future works 655.1 Conclusions . . . . . . . . . . . . . . . . . . . . 655.2 Future Works . . . . . . . . . . . . . . . . . . . 66Bibliography 67List of Figures1.1 Types of medical i

maging. . . . . . . . . . . . . . 21.2 Supervised and Unsupervised Learning algorithms based on tasks. . . . . . 61.3 Various architectures based on computer vision tasks. . . . . . . . . . . . . 81.4 Healthy kidneys. . . . . . . . . . . . . . . . . . 121.5 ADPKD kidneys on NCCT and CCT, and renal

cyst: (a) ADPKD kidneyand liver cyst on CCT; (b) ADPKD kidney and liver cyst on NCCT; (c)ADPKD kidney, liver, and spleen; (d) Renal cyst in non-ADPKD. . . . . . 143.1 Raw image and respective ground truth images: (a) Raw image; (b) Groundtruth for right kidney (Green); (c) Ground truth for left kidn

ey (Yellow); (d)Ground truth for both right (Green) and left (Yellow) kidneys. . . . . . . . 253.2 Proposed automatic ADPKD kidneys localization model framework. . . . . 263.3 Preprocessing procedures. .. . . . . . . . . . . . 263.4 The architecture of automatic ADPKD kidneys localization model, whe

rejfj, C(x, y), w, h, and V2 refer to as total number of feature maps, centerbounding box, width, height, and version 2, respectively. . . . . . . . . . . 293.5 Precision and recall curve on NCCT: (a) Right kidney; (b) Left kidney. . . . 363.6 Automatic ADPKD kidneys localization results on NCCT. .

. . . . . . . . 363.7 Precision and recall curve on CCT: (a) Right kidney; (b) Left kidney. . . . . 393.8 Automatic ADPKD kidneys localization results on CCT. . . . . . . . . . . 393.9 Precision and recall curve of our model on image-wise testing set: (a) Rightkidney; (b) Left kidney. . . . . . . .

. . . . . . . 41xiv3.10 Comparison of classification and localization loss on image-wise testing set. 423.11 Detection results: (a) ADPKD kidneys associated with liver cysts; (b)ADPKD kidneys with adjacent organs. . . . . . . . . . 423.12 Precision and recall curve of our model on subject-wise test

ing: (a) Rightkidney; (b) Left kidney. . . . . . . . . . . . . . .. 433.13 Comparison of classification and localization loss on image-wise testing set. 443.14 Detection results: (a) Small size of ADPKD kidneys; (b) Big size of ADPKDkidneys. . . . . . . . . . . . . . . . . . . . . . .. 453.15 Miscla

ssification and mislocalization example: (a) Misclassification; (b)Mislocalization. . . . . . . . . . . . . . . . . . .. 464.1 The overview of proposed method work flow. . . .. 504.2 The overview of data preprocessing. . . . . . . . 514.3 The overview of automatic ADPKD kidneys segmentation model. .

. . . . 534.4 Automatic ADPKD kidneys segmentation ROC curve on NCCT. . . . . . . 604.5 Automatic ADPKD kidneys segmentation results on NCCT. . . . . . . . . 614.6 Automatic ADPKD kidneys segmentation ROC curve on CCT. . . . . . . . 614.7 Automatic ADPKD kidneys segmentation results on CCT. . . . .

. . . . . 624.8 Validation curve for DTR: (a) Validation score on NCCT; (b) Validationscore on CCT. . . . . . . . . . . . . . . . . . . . 63List of Tables3.1 210 CT data acquisitions from 97 ADPKD patients. . . 243.2 Localization results using validation set with k-fold on NCCT. . . . . . . . 343.3

Localization results using testing set on NCCT. . . .353.4 Localization results using validation set with k-fold on CCT. . . . . . . . . 373.5 Localization results using testing set on CCT. . . . 383.6 Evaluation metrics results on image-wise testing. . .403.7 Comparison of AP and mAP with other ar

chitectures on image-wise testing. 413.8 Evaluation metrics results of subject-wise testing. .433.9 Comparison of AP and mAP with other architectures on subject-wise testing. 444.1 Segmentation results using validation set with k-fold on NCCT. . . . . . . . 584.2 Segmentation results using testing s

et on NCCT. . . .594.3 Segmentation results using validation set with k-fold on CCT. . . . . . . . . 624.4 Segmentation results using testing set on CCT. . . . 634.5 R2 score using validation set with k-fold on NCCT and CCT. . . . . . . . . 64