single-image-super-resolution-1032

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 achitecture. It enhances the resolution of the input image by a factor of 4.

Example

Low resolution:

street_480x270.png

Bicubic interpolation:

x4c_street_480x270.png

Super resolution:

x4_street_480x270.png

Specification

Metric Value
PSNR 29.29 dB
GFlops 11.654
MParams 0.030
Source framework PyTorch*

For reference, PSNR for bicubic upsampling on test dataset is 26.79 dB.

Performance

Inputs

  1. name: "0" , shape: [1x3x270x480] - An input image in the format [BxCxHxW], where:
    • B - batch size
    • C - number of channels
    • H - image height
    • W - image width.
  2. name: "1" , shape: [1x3x1080x1920] - Bicubic interpolation of the input image in the format [BxCxHxW], where:

    • B - batch size
    • C - number of channels
    • H - image height
    • W - image width.

    Expected color order is BGR.

Outputs

  1. The net outputs one blobs with shapes [1, 3, 1080, 1920] that contains image after super resolution.

Legal Information

[*] Other names and brands may be claimed as the property of others.