APL in PyTorch

Implementation of "United Defocus Blur Detection and Deblurring via Adversarial Promoting Learning" in PyTorch

Datasets

• train_data:

  • 1204source: Contains 604 training images of CUHK Dataset and 600 training images of DUT Dataset.
  • FC: 500 natural full clear images.

• test_data of DBD:

  • CUHK: Contains 100 testing images of CUHK Dataset and it's GT.
  • DUT: Contains 500 testing images of DUT Dataset and it's GT.

• test_data of Deblurring:

  • CUHK: Contains 100 testing images.
  • DUT: Contains 500 testing images.
  • DP: Contains 76 testing images of DP Dataset and it's GT.

Download and unzip datasets from https://github.com/shangcai1/SG [1] to "./dataset". Add testset of DP datasets from https://ln2.sync.com/dl/c45358c50/r7kpybwk-xw8hhszh-qkj249ap-y8k2344d/view/default/10770664840008 [2] to "./dataset/test/DP".

Test

You can use the following command to test：

python test.py --image_path TEST_DATA_PATH --result_save_path RESULT_IMAGE_PATH

You can use the following model to output results directly.

Here are our parameters:

baidu link: https://pan.baidu.com/s/1sAbhPioPCLrsAid1W8UMAg?pwd=t8hq password: t8hq

google drive: https://drive.google.com/drive/folders/1lPcoIY-lqKAvsfvuYUgB9mmM6YKBy2h8?usp=sharing

Put "DBD.pth" and "deblur.pth" in "./saved_models".

Train

You can use the following command to train：

python train.py --data_root TRAIN_DATA_PATH

• train.py: the entry point for training.
• models/our_model.py: the whole model of APL.
• models/DBDNet.py: defines the architecture of the DBD Generator models.
• models/DeblurNet.py: defines the architecture of the Deblur Generator models and Discriminator models.
• options.py: creates option lists using argparse package. More individuals are dynamically added in other files as well.
• datasets.py: process the dataset before passing to the network.
• optimizer.py: defines the optimization and losses used in APL.

Eval

DBD

If you want to use Fmax and MAE to evaluate the results, you can run the following code in MATLAB. It shows the PR curve and F-measure curve at the same time.

./evaluate_dbd/evaluate.m

Deblurring

If you want to use PSNR, SSIM and MAE to evaluate the result, use the following code:

python evaluate.py --image_save_path RESULT_IMAGE_PATH --test_gt_path GT_PATH

References

<a id="1">[1]</a> Zhao, W., Shang, C., Lu, H.: Self-generated defocus blur detection via dual adversarial discriminators. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.

<a id="2">[2]</a> Abuolaim, A., Brown, M.S.: Defocus deblurring using dual-pixel data. In: European Conference on Computer Vision.