The reason why super-scoring is necessary is because the corner points of the checkerboard are not detected well. When the picture is super-scored, the checkerboard actually becomes larger because there are more pixels.
This is the effect of detection.
This is probably the effect. But the annoying thing is that the ultra-high-resolution model is too big. This is a cropped picture that can be super-high-resolution, but it won’t work if it’s bigger. It’s really fucked. ISR code library
In order to calibrate the camera, using super score is overkill.
It is better to take two clear and larger photos.
other papers and code
Paper summary:
An image super-resolution code written in Lua. The example given by the author is a two-dimensional image. I don’t know if it is suitable for ordinary images. Link to
the ISR code base.
The above is too big in Linma and takes up video memory. I found a small
paddleGAN
super-resolution tutorial.