Openslide: select best input level based on output dimensions

[TAYTS]

I am trying to generate thumbnail file from a large openslide format image(~2Gb), but the resize takes forever to run, not sure if I did it wrongly. Below are my code for trying to generate the thumbnails.

I manage to use vipsthumbnail to generate the thumbnail very fast(instant) and I am using the global libvips.

sharp('test.scn', {limitInputPixels: false})
  .resize(500, 500)
  .png()
  .toFile('output.png')
  .then(info => console.log(info))
  .catch(err => console.log(err))

[lovell]

Are you able to provide a sample image?

What versions of sharp, libvips and openslide are you using? On what platform?

[kleisauke]

Note that vipsthumbnail uses shrink-on-load for Openslide images since version 8.8, which might explain the speed improvement you noticed. See:
libvips/libvips@22ba910
libvips/libvips#1149

[TAYTS]

@lovell Sorry for missing out the details.

sharp: 0.25.3
libvips: 8.9.2
openslide: 3.4.1
platform: macOS

I am afraid I am not able to provide the actual image but here is a link to a sample data:
http://openslide.cs.cmu.edu/download/openslide-testdata/Leica/Leica-3.scn

[TAYTS]

@kleisauke Is the shrink-on-load currently supported by sharp or I have missed out the resize configuration as the libvips version stated in the package.json is 8.9.1?

[TAYTS]

Is the fast thumbnail generation supported in the sharp as I can't seem to find the documentation about it?

[lovell]

Commit f5f1654 on the work-in-progress openslide-shrink-on-load branch attempts to add this feature.

@TAYTS Please can you test this.

npm install lovell/sharp#openslide-shrink-on-load

[TAYTS]

@lovell I got [Error: webp: unable to parse image] after using the new version even with clean install. Do I need to configure the file type specifically? I am testing using the sample data from http://openslide.cs.cmu.edu/download/openslide-testdata/Leica/

[lovell]

I think the Leica-3.scn sample fails due to openslide/openslide#189 - or are you referring to a different image?

[TAYTS]

@lovell I have tried on the smaller image Leica-1.scn and it works on the current stable version (able to generate the output), but I get the same error (unable to parse image) for the new version.

[lovell]

Thank you. It looks like shrink-on-load for openslide input is a little more complex than providing a scaling factor. We need to calculate the best level to extract, based on its dimensions vs desired dimensions. Here's the relevant metadata from the Leica-1.scn sample:

openslide.level-count: 5
openslide.level[0].downsample: 1
openslide.level[0].height: 153470
openslide.level[0].width: 53130
openslide.level[1].downsample: 3.9998986524159981
openslide.level[1].height: 38368
openslide.level[1].width: 13283
openslide.level[2].downsample: 15.998992404088622
openslide.level[2].height: 9592
openslide.level[2].width: 3321
openslide.level[3].downsample: 63.927109191868013
openslide.level[3].height: 2401
openslide.level[3].width: 831
openslide.level[4].downsample: 255.3952147062588
openslide.level[4].height: 601
openslide.level[4].width: 208

Let's tag this as an enhancement for now.

[lovell]

The "quickest" (to implement) solution to help with this is to allow the extraction of a specific level, rather than the current behaviour of always using the first/largest layer.

The first step in commit 760550c exposes levels via metadata, e.g. for the Leica-1.scn example:

{
  format: 'openslide',
  width: 53130,
  height: 153470,
  space: 'srgb',
  channels: 4,
  depth: 'uchar',
  isProgressive: false,
  levels: [
    { width: 53130, height: 153470 },
    { width: 13283, height: 38368 },
    { width: 3321, height: 9592 },
    { width: 831, height: 2401 },
    { width: 208, height: 601 }
  ],
  hasProfile: false,
  hasAlpha: true
}

The next step is to allow the required level to be specified in the sharp() constructor options.

It will then at least be possible for someone to select the most relevant level based on their desired output dimensions.

[lovell]

Commit 7f142bd adds the level constructor option to use a specific level.

sharp('Leica-1.scn', {
  level: 4,
  limitInputPixels: false
})
...

This will be in v0.25.4.

[lovell]

v0.25.4 is now available.