Upscale low-resolution images using sub-pixel convolutions (ESPCN)
Uses the ESPCN (Efficient Sub-Pixel Convolutional Network) architecture with Conv2d layers and PixelShuffle for learnable upscaling. The network learns to reconstruct high-frequency details from low-resolution input. PSNR (Peak Signal-to-Noise Ratio) measures reconstruction quality in decibels.
