How pratik5200/vlm-project works
This repository is a notebook-based image-restoration project. Its documented main workflow detects image degradations, runs DA-CLIP restoration for each detected type, combines the outputs, and evaluates the result. It also explores AutoDIR as a related framework and describes DIV2K data preparation. The notebook contents were not provided, so the diagram reflects README-documented stages rather than verified implementation details or notebook-specific responsibilities.
Main entry points: da-clip-updated.ipynb Quality evaluation · autodir .ipynb Degradation classification · daclip .ipynb Data preparation
Components 13
Restoration workflow
- Degraded image
- Degradation finder
da-clip-updated.ipynb - DA-CLIP restoration
da-clip-updated.ipynb - Result combining
da-clip-updated.ipynb - Restored image
Data and evaluation
- DIV2K data
- Data preparation
daclip .ipynb - Quality evaluation
da-clip-updated.ipynb - PSNR SSIM LPIPS
- Saved results
results/Result_output.zip
AutoDIR exploration
- Degradation classification
autodir .ipynb - Latent diffusion restoration
autodir .ipynb
Other components
- Image user
Connections 10
- Image user → Degraded image: provides
README.md - Degraded image → Degradation finder: detects types
README.md - Degradation finder → DA-CLIP restoration: conditions passes
README.md - DA-CLIP restoration → Result combining: produces outputs
README.md - Result combining → Restored image: merges into
README.md - DIV2K data → Data preparation: provides images
README.md - Restored image → Quality evaluation: compares with reference
README.md - DIV2K data → Quality evaluation: provides reference
README.md - Quality evaluation → PSNR SSIM LPIPS: reports
README.md - Degradation classification → Latent diffusion restoration: detects then restores
README.md