Separating Luminance (Y) from Chrominance (U, V) to drastically reduce neural processing cost.
In this POC, we convert incoming RGB to YUV (Rec. 709). The target for the heavy 6-layer CNN is only the Y channel, where most high-frequency detail resides.
Channel Roles:
• Y (Luminance): High-res neural upscaling target.
• U/V (Chrominance): Lightweight bicubic upscaling target.