This Repository is my assignment for cs3324 on IQA model design, based on chaofengc/IQA-PyTorch: 👁️ 🖼️ 🔥PyTorch Toolbox for Image Quality Assessment, including LPIPS, FID, NIQE, NRQM(Ma), MUSIQ, NIMA, DBCNN, WaDIQaM, BRISQUE, PI and more... (github.com)
# Install the requirements, a python 3.8 environment is recommended.
pip install -r requirements.txt
# As this repository is basically my modified version of pyiqa, you need to first uninstall the original pyiqa.
pip unintall pyiqa
# Then install my version.
python setup.py develop
Install torch form PyTorch if you want to use cuda.
Now the structure should be like:
your_cloned_repo/
|
|___ datasets/
| |
| |___ AGIQA-3K/
| | |
| | |___ image1.jpg ...
| |
| |___ other_datasets/
| |
| |___ meta_info/
| |
| |___ AGIQA-3K.pkl
| |
| |___ meta_info_AGIQA-3K.csv
|
|___ other_dirs/
You can directly click on './start.bat', if you are using Windows and exactly one GPU.
Or run by command:
# In the env you set up previously:
python pyiqa/train.py -opt options/train/train_AGICQ-3K_myiqa.yml
If the training process fails, try to adjust the parameters 'num_worker_per_gpu' and 'batch_size_per_gpu' in the settings file '.\options\train\train_AGICQ-3K_myiqa.yml'.
Original results stored in './myresults', including logs and '.pth' file of SVD_CNNIQA.