# deblurgan-v2¶

## Use Case and High-Level Description¶

DeblurGAN-v2 is a generative adversarial network (GAN) for single image motion deblurring. This model is based on a relativistic conditional GAN with a double-scale discriminator. For details about architecture of model, check out the paper. Model used MobileNet as backbone and was trained on GoPro, DVD, NFS datasets. For details about implementation of model, check out the DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better repository.

Metric

Value

Type

Image Processing

GFLOPs

80.8919

MParams

2.1083

Source framework

PyTorch*

## Accuracy¶

Model was tested on GoPro test dataset.

Metric

Original model

Converted model

PSNR

28.25Db

28.24Db

SSIM

0.97

0.97

## Input¶

### Original Model¶

Image, name - blur_image, shape - 1, 3, 736, 1312, format is B, C, H, W, where:

• B - batch size

• C - channel

• H - height

• W - width

Channel order is RGB. Mean values - [127.5, 127.5, 127.5], scale values - [127.5, 127.5, 127.5].

### Converted Model¶

Image, name - blur_image, shape - 1, 3, 736, 1312, format is B, C, H, W, where:

• B - batch size

• C - channel

• H - height

• W - width

Channel order is BGR.

## Output¶

### Original Model¶

Deblurred image, name - deblur_image, shape - 1, 3, 736, 1312, output data format is B, C, H, W, where:

• B - batch size

• C - channel

• H - height

• W - width

Channel order is RGB.

### Converted Model¶

Deblurred image, name - deblur_image, shape - 1, 3, 736, 1312, output data format is B, C, H, W, where:

• B - batch size

• C - channel

• H - height

• W - width

Channel order is BGR.

You can download models and if necessary convert them into OpenVINO™ IR format using the Model Downloader and other automation tools as shown in the examples below.

omz_downloader --name <model_name>
omz_converter --name <model_name>