iPAS AI 應用規劃師中級 114 年第二次 科三 第 43 題 詳解

VGG16是由牛津大學Visual Geometry Group(VGG)在2014年提出的經典卷積神經網路（Convolutional Neural Network, CNN）架構。該模型以簡潔且規則的層堆疊設計聞名，廣泛應用於影像分類、特徵提取及遷移學習等任務。附圖程式碼載入了預訓練的VGG16模型，並輸出其完整層級結構及參數統計摘要（如附表）。請根據此資訊回答42~45題。 ---------------------------------------------------------------- Layer (type) Output Shape Param # ================================================================ Conv2d-1 [-1, 64, 150, 150] 1,792 ReLU-2 [-1, 64, 150, 150] 0 Conv2d-3 [-1, 64, 150, 150] 36,928 ReLU-4 [-1, 64, 150, 150] 0 MaxPool2d-5 [-1, 64, 75, 75] 0 Conv2d-6 [-1, 128, 75, 75] 73,856 ReLU-7 [-1, 128, 75, 75] 0 Conv2d-8 [-1, 128, 75, 75] 147,584 ReLU-9 [-1, 128, 75, 75] 0 MaxPool2d-10 [-1, 128, 37, 37] 0 Conv2d-11 [-1, 256, 37, 37] 295,168 ReLU-12 [-1, 256, 37, 37] 0 Conv2d-13 [-1, 256, 37, 37] 590,080 ReLU-14 [-1, 256, 37, 37] 0 Conv2d-15 [-1, 256, 37, 37] 590,080 ReLU-16 [-1, 256, 37, 37] 0 MaxPool2d-17 [-1, 256, 18, 18] 0 Conv2d-18 [-1, 512, 18, 18] 1,180,160 ReLU-19 [-1, 512, 18, 18] 0 Conv2d-20 [-1, 512, 18, 18] 2,359,808 ReLU-21 [-1, 512, 18, 18] 0 Conv2d-22 [-1, 512, 18, 18] 2,359,808 ReLU-23 [-1, 512, 18, 18] 0 MaxPool2d-24 [-1, 512, 9, 9] 0 Conv2d-25 [-1, 512, 9, 9] 2,359,808 ReLU-26 [-1, 512, 9, 9] 0 Conv2d-27 [-1, 512, 9, 9] 2,359,808 ReLU-28 [-1, 512, 9, 9] 0 Conv2d-29 [-1, 512, 9, 9] 2,359,808 ReLU-30 [-1, 512, 9, 9] 0 MaxPool2d-31 [-1, 512, 4, 4] 0 AdaptiveAvgPool2d-32 [-1, 512, 7, 7] 0 Linear-33 [-1, 4096] 102,764,544 ReLU-34 [-1, 4096] 0 Dropout-35 [-1, 4096] 0 Linear-36 [-1, 4096] 16,781,312 ReLU-37 [-1, 4096] 0 Dropout-38 [-1, 4096] 0 Linear-39 [-1, 1000] 4,097,000 ================================================================ Total params: 138,357,544 Trainable params: 138,357,544 Non-trainable params: 0 ---------------------------------------------------------------- Input size (MB): 0.26 Forward/backward pass size (MB): 96.93 Params size (MB): 527.79 Estimated Total Size (MB): 624.98 ----------------------------------------------------------------