liyucs / dpanet

The implementation of "Learning Progressive Dual-Pixel Alignment for Defocus Deblurring".

DPANet

The implementation of "Learning Dual-Pixel Alignment for Defocus Deblurring".

Prerequisites

• The code has been tested with the following environment
  • Ubuntu 18.04
  • Python 3.7.9
  • PyTorch 1.7.0
  • cudatoolkit 10.0.130
  • NVIDIA TITAN RTX GPU

Datasets

Training datasets

• DPDD train set

Testing datasets

• DPDD test set
• PIXEL test set

Preparation (for DCNv2)

$ cd DPANet
$ python setup.py build develop

Test

• Download our pre-trained model and put the final.pth into ./checkpoint folder

$ cd DPANet
$ python test.py

For more results, you can refer to More results on DPDD.

Train

• First, crop the images of DPDD train set into 512*512 patches using the same settings as DPDNet. (You can use $ python ./image_to_patch_filter.pyfrom DPDD to get the patches.)
• After getting the training patches, please organize the training dataset according to our code implementation: $ mkdir dpdd_datasets/dpdd_16bit, then move the train/test folder into the directory just created.

Start training

$ cd DPANet
$ python train.py

During training, we first train DPANet with MSE loss. After that, we choose the checkpoint that gives the best result among all the epochs (about 300 epochs, example ckpt trained with MSE) and finetune it with Charbonnier loss.

Results

Here we give results of different methods on DPDD and PIXEL datasets.

• DPDD results
• PIXEL results

dpanet's Issues

cropping patch for training

In DPANet，the method of getting training image patches is differen with DPDD.
What are the advantages of your processing?

Learning Dual-Pixel Alignment for Defocus Deblurring

Thanks for your kindly share the code, Can you share the paper?

Gets some problems when I run python setup.py build develop

Hello, Dear sir,

Thanks your for you kindly sharing your code.
And I actually meet some error "AssertionError: Torch not compiled with CUDA enabled" when I run the command lines
$ cd DPANet
$ python setup.py build develop.

I think it might be caused by the version of CUDA in my PC (my cudatools or something related to CUDA is much higher than cudatoolkits 10.2). Could please you give me some suggestions about this error? And I would like to ask why do you use the compiled code of DCN2 but not use the code of DCN2 which is no need to compile the CPP?

Best regards