Implementing a 6-layer Convolutional Neural Network specifically for the Luminance (Y) channel.
Architecture:
• Layer 1: Conv2D 3x3 (32 filters)
• Layer 2-5: Depthwise Conv (32 filters)
• Layer 6: Conv2D 3x3 (1 filter output)
Sub-pixel Reconstruction: The final layer generates 4 pixels for every 1 input pixel (arranged in 2x2 grid) to perform 2x super-resolution without checkerboard artifacts.