深度神经网络架构图

一张从左到右的深度卷积神经网络架构图，计算机视觉论文风格。输入为左端 32×32×3 像素网格图标；依次展示天蓝色 Conv2D 立方体（「Input 32×32×3」）、珊瑚色 ReLU 块、哑金色 MaxPool 立方体，重复两次（128、256 滤波器，立方体渐宽）；两个青绿色 FC 全连接矩形（「Conv 3×3, 64」 和 「ReLU」）；最终鼠尾草绿 Softmax 输出块（10 类圆形节点列）。层间细箭头，各块上方维度标注（如 "16×16×64"）；Conv 与 ReLU 间细石板灰批归一化带；FC 层带小叉圆 Dropout 图标。白色背景，竖向居中，9pt Helvetica 字体。

A clean vector diagram of a deep convolutional neural network architecture, presented horizontally from left to right in the style of an academic computer vision paper. The network processes a small input image (32 × 32 pixel grid icon at the far left, labelled "Input 32×32×3"). The architecture flows right: Conv2D layer shown as a flat 3D cuboid in sky blue (labelled "Conv 3×3, 64"), followed by a ReLU activation block in coral ("ReLU"), then a MaxPooling cuboid in muted gold ("MaxPool 2×2"), repeated twice with increasing depth (128, 256 filters) shown by progressively wider cuboids. Then two Fully Connected (FC) layers as narrow tall rectangles in teal, the first labelled "FC 512" and the second "FC 128". Finally, a Softmax output block in sage green with 10 output nodes in a small circle column labelled "Softmax 10 classes". Each layer is connected by thin horizontal arrows showing data flow. Dimension annotations above each block (e.g., "16×16×64"). Batch normalisation shown as a thin slate-grey band between Conv and ReLU blocks. Dropout indicated by small crossed-out circle icons on FC layers. Background: white. Layout: left-to-right, centre-aligned vertically. Font: 9 pt Helvetica-style.