预览提示词
Make a horizontal technical architecture diagram of a convolutional neural network image-classification pipeline, drawn as a left-to-right roadmap of stacked block stages connected by directional arrows. Start at the left with an input RGB image labelled "224x224x3", then flow through repeated Conv 3x3 + ReLU + MaxPool 2x2 stages drawn as 3D slab stacks whose feature maps visibly shrink in spatial size while growing in channel depth; annotate dimensions at each stage (112x112x32, 56x56x64, 28x28x128, 14x14x256). Show two small feature-map activation visualizations (early edge maps, deeper texture maps) as inset thumbnails above their layers. Then a Flatten step into two Fully-Connected layers labelled "FC 512" and "FC 128", ending in a Softmax output drawn as a horizontal class-probability bar chart with classes Cat 0.86, Dog 0.09, Bird 0.03, Other 0.02. Add a thin parameter-count strip beneath. Keep it clean and publication-quality with tidy alignment, legible labels, crisp monospaced dimension tags, and a pure-white background.
