PyTorch 0.4 implementation of the following paper: Kang L, Ye P, Li Y, et al. Convolutional neural networks for no-reference image quality assessment[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2014: 1733-1740.
The optimizer is chosen as Adam here, instead of the SGD with momentum in the paper.
CUDA_VISIBLE_DEVICES=0 python main.py --exp_id=0 --database=LIVE
Before training, the im_dir
in config.yaml
must to be specified.
Train/Val/Test split ratio in intra-database experiments can be set in config.yaml
(default is 0.6/0.2/0.2).
Test Demo
python test_demo.py --im_path=data/I03_01_1.bmp
python test_cross_dataset.py --help
TODO: add metrics calculation. SROCC, KROCC can be easily get. PLCC, RMSE, MAE, OR should be calculated after a non-linear fitting since the quality score ranges are not the same across different IQA datasets.
tensorboard --logdir=tensorboard_logs --port=6006
- PyTorch 0.4
- TensorboardX 1.2, TensorFlow-TensorBoard
- pytorch/ignite