ImageRestoration

Image sizes:

1. June Brewer 644x773 96dpi 32bit
2. Wilhelmina Perry 876x944 144dpi 32bit
3. Redd Dudley 1351x1656 600dpi 24bit
4. Arleen Lawson 1220x1746 600dpi 32bit ##Day 2 (11.29.23) ###Planning

• Meet with grad student whenever available

1. Consolidate results and narrow down models
2. Run 4 sample photos with all models
   • convert tif files (higher quality?) to png

Models/Tools

• fotor.com
• ILO (cuda required)
• Microsoft
• SuperSR (cuda required)
• PSLD
• ancestry
• hotpot.ai Additional Considerations
• Conversions to jpeg inputs may work better than png on some models ##Day 1 (11.28.23)

• Fotor is surprisingly good (but tends to apply makeup)
• Microsoft is good for scratches
• PLSD is bad ###Planning Potential Models:

1. ILO (https://github.com/giannisdaras/ilo) (Anna, Luke)
2. PLSD (Arnold)
3. SuperSR (Isshan)
4. Microsoft (Raymond)
5. HuggingFace (Brandon) Evaluation:
6. Manual Check
7. Structural Similarity Index (SSI)
8. Peak Signal-to-Noise Ratio (PSNR)
9. Visual Information Fidelity (VIF)
10. Crowd Sourcing/Polling

MPRNet from PIL import version as PILLOW_VERSION in /home/ayg479/opt/anaconda3/envs/pytorch1/lib/python3.7/site-packages/torchvision/transforms/functional.py