← all work generative-vision-lab

Face restoration

Restore degraded face crops. Degradation is synthesised at load time.

Result

splitPSNR
train n=15030.0851
val n=15029.9772
test n=5030.0195

The finding

Notes

- Generator fine-tuned with L1 on synthetically degraded FFHQ crops, 15,000 steps at batch 2. - Sample images show the degraded input, the pretrained output and the fine-tuned output against ground truth.

Figures

00526_after_output.png
00526_after_output
00526_before_output.png
00526_before_output
00526_ground_truth.png
00526_ground_truth
00526_input_degraded.png
00526_input_degraded
00732_after_output.png
00732_after_output
00732_before_output.png
00732_before_output
00732_ground_truth.png
00732_ground_truth
00732_input_degraded.png
00732_input_degraded
00795_after_output.png
00795_after_output
00795_before_output.png
00795_before_output
00795_ground_truth.png
00795_ground_truth
00795_input_degraded.png
00795_input_degraded

Elsewhere

Code and full results on GitHub

Every number is read from the file the run produced. github.com/hammasbuilds · kaggle.com/muhammadhammas13