Evaluated images cannot always be compared with reference images. There is often a situation of lack of reference images. In such cases, we evaluate the image based on, for example, sharpness, color and contrast. For such evaluation, deep neural networks can also be used.
- Python 3.10
- TensorFlow 2.15
- other, listed in requirements.txt
For training and evaluation LIVE and TID2013 are used.
- Run
preprocessing.m
inside the directory where the LIVE dataset is stored. - Import the distorted_images folder into your working directory.
- Run the CNNIQA or CNNIQA++ notebook. You can choose to use the dataset directly or perform cross-validation (use one part for training and the other for testing).
TensorFlow implementations of the following papers: