single-image-super-resolution-1033

Use Case and High-Level Description

An Attention-Based Approach for Single Image Super Resolution but with reduced number of channels and changes in network architecture. It enhances the resolution of the input image by a factor of 3.

Example

Low resolution:

_images/street_640x360.png

Bicubic interpolation:

_images/x3c_street_640x360.png

Super resolution:

_images/x3_street_640x360.png

Specification

Metric

Value

PSNR

30.97 dB

GFlops

16.062

MParams

0.030

Source framework

PyTorch*

For reference, PSNR for bicubic upsampling on test dataset is 28.48 dB.

Inputs

  1. Image, name: 0, shape: 1, 3, 360, 640 in the format B, C, H, W, where:

    • B - batch size

    • C - number of channels

    • H - image height

    • W - image width

  2. Bicubic interpolation of the input image, name: 1, shape: 1, 3, 1080, 1920 in the format B, C, H, W, where:

    • B - batch size

    • C - number of channels

    • H - image height

    • W - image width

    Expected color order is BGR.

Outputs

The net output is a blob with shapes 1, 3, 1080, 1920 that contains image after super resolution.

Demo usage

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities: