Comparison of image reconstruction metrics across varying downsampling factors α = 8, 16, 64 using a VQGAN-based encoder-decoder architecture. The metrics include: (a) LPIPS to assess perceptual similarity, (b) PSNR for signal fidelity, (c) RMSE for pixel-wise error, and (d) SSIM for structural similarity. There is a noticeable trade-off between compression and image quality, with lower values of α resulting in better perceptual and structural performance, while higher α values lead to increased error and degradation in perceptual metrics.