Installation

conda create -n agenticir python=3.10
conda activate agenticir
pip install -r installation/requirements.txt

We employ the following models as single-degradation restoration tools: DiffBIR, X-Restormer, SwinIR, HAT, MPRNet, MAXIM, Restormer, DRBNet, IFAN, RIDCP, DehazeFormer, and FBCNN. Thanks for these awesome works.

Run sh installation/deploy_tools.sh to prepare the code, which is adapted from the official repos linked above.
Set up their respective environments according to the official repos.
Download the weights. You may need to modify the paths in files in executor. A tutorial will be given if necessary.
Run python -m test_tool.test_tool to check whether all tools work properly.

Run sh installation/deploy_depictqa.sh to prepare the code, which is adapted from the official repo linked above.
Set up the environment according to the official repo.
Download the weights.
- Download the pre-trained ViT from this link and put it in DepictQA/weights/.
- Download the pre-trained Vicuna from this link and put it in DepictQA/weights/.
- Download the delta weights of DepictQA-Wild from this link, rename it to DQ495K.pt, and put it in DepictQA/weights/delta/.
- Download the delta weights fine-tuned from DepictQA-Wild from this link and put it in DepictQA/weights/delta/.
The structure of DepictQA/weights should look like this:
```
DepictQA/weights/
├── ViT-L-14.pt
├── vicuna-7b-v1.5/
│   └── ...
└── delta/
    ├── DQ495K.pt
    └── degra_eval.pt
```

Provide feedback