프롬프트 설명

A clean and professional neural network architecture diagram illustrating a lightweight encoder–decoder semantic segmentation model. The encoder is a MobileNetV2 backbone producing multi-scale feature maps. The decoder starts from the lowest-resolution encoder output, applies a 1×1 convolution for channel reduction, followed by progressive bilinear upsampling. At each upsampling stage, skip connections fuse intermediate encoder feature maps with decoder features. The decoder uses only 1×1 convolutions and bilinear upsampling, without heavy modules such as ASPP or attention. The final output is a full-resolution segmentation map. Flat vector-style, clear arrows, labeled blocks, academic paper figure style, white background.