Hi I am trying to use vizarr with <a class="user-mention notranslate" data-hovercard-t

Hi <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="

<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/us

All works now, see a gif below: <a target="_blank" rel="noopener noreferrer nofoll

View zarr loaded with napari_lazy_openslide about vizarr HOT 6 CLOSED

hms-dbmi commented on June 13, 2024

View zarr loaded with napari_lazy_openslide

from vizarr.

Comments (6)

manzt commented on June 13, 2024 2

Wow, cool!

How do I make it display in color mode, I guess that has to do with the "omero" key, right? What's the best way to pass that?

Omero key shouldn't be necessary. I believe if you set channel_axis = 0 the colors should work.

Another minor issue is that, if I zoom in too much it becomes blank, can I limit the zoom level?

Hmm I'm not sure about this. I have a PR I'm about to merge upgrading Viv (#126), and I wonder if that would vix the issue. In the run_vizarr function, you can set the url to the netlify deploy preview link and see if the upgrade fixes it.

BTW, would you consider having such feature in napari_lazy_openslide so it can work out of the box?

Yeah, definitely. I wonder if this functionality would be best implemented as a method on the OpenSideStore that returns another class wrapping the store (instead of inheriting the store). E.g.

from napari_lazy_openslide import OpenSlideStore

OpenSlideStore('./data.svs').as_ngff() # returns store transformer

EDIT: Also, @oeway, an as_ngff() method could add the additional axes metadata to the mulitscales metadata so that channel_axis does not need to be provided.

from vizarr.

manzt commented on June 13, 2024

Hi @oeway ! We currently don't support interleaved RGB/A images in the vizarr client. The last two dimensions of an array must be YX. However, we have added support for this in Viv, and I can look into adding that functionality client-side in vizarr.

I'm sorry, but as a work around on the python side, you could wrap the OpenslideStore in another zarr store that performs the translation of (y, x, c) to (c, y, x). You need to proxy requests for .zarray metadata to make the following changes:

shape: (y, x, 4) -> (4, y, x)
chunks (tilesize, tilesize, 4) -> (1, tilesize, tilesize).

Any zarr client will now request keys in the form "level/c.y.x" from the store. It gets tricky because the chunk shapes are different but in theory (I haven't tested this) it could look like:

from napari_lazy_openslide import OpenSlideStore
from zarr.util import json_dumps, json_loads
import numpy as np

def _parse_chunk_path(path: str):
    """Returns x,y chunk coords and pyramid level from string key"""
    level, ckey = path.split("/")
    c, y, x  = map(int, ckey.split("."))
    return c, x, y, int(level)

class Wrapper(OpenSlideStore):

  def __getitem__(self, key):
    if key in self._store: # metadata store for OpenSlideStore
       # key is for metadata
       res = self._store[key]
       if key.endswith('.zarray'):
         meta = json_loads(res)
         meta.shape = [meta.shape[2], meta.shape[0], meta.shape[1]]
         meta.chunks = [1, meta.chunks[0], meta.chunks[1]]
         return json_dumps(meta)
       else:
         # zgroup
         return res

      cidx, y, x, level = _parse_chunk_path(key)
      buf = store[f"{level}/{y}.{x}.0"] # could cache this result
      arr = np.frombuffer(buf, dtype='uint8').reshape(self.tilesize, self.tilesize, -1)
      data = arr[:, :, cidx] # cou
      return data.tobytes()

It gets a little tricky because the chunk shape of the original array is (tilesize, tilesize, 4) and the zarr array would need to map to (1, tilesize, tilesize).

from vizarr.

oeway commented on June 13, 2024

Hi @manzt Thanks for the detailed explanation!

I started to try the Wrapper code you wrote above, and here is a version with some fix, but still not working:

from napari_lazy_openslide import OpenSlideStore
from zarr.util import json_dumps, json_loads
import numpy as np

def _parse_chunk_path(path: str):
    """Returns x,y chunk coords and pyramid level from string key"""
    level, ckey = path.split("/")
    c, y, x  = map(int, ckey.split("."))
    return c, x, y, int(level)

class Wrapper(OpenSlideStore):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

    def __getitem__(self, key):
        if key in self._store: # metadata store for OpenSlideStore
            # key is for metadata
            res = self._store[key]
        print(self._store.keys())
        if key.endswith('.zarray'):
            meta = json_loads(res)
            meta["shape"] = [meta["shape"][2], meta["shape"][0], meta["shape"][1]]
            meta["chunks"] = [1, meta["chunks"][0], meta["chunks"][1]]
            return json_dumps(meta)
        elif key.endswith('.zgroup') or key.endswith('.zattrs'):
            # zgroup
            return res
        
        cidx, y, x, level = _parse_chunk_path(key)
        buf = self._store[f"{level}/{y}.{x}.0"] # <-------------------------Is this correct?
        arr = np.frombuffer(buf, dtype='uint8').reshape(self.tilesize, self.tilesize, -1)
        data = arr[:, :, cidx] # cou
        return data.tobytes()

You wrote store[...] in your previous code, I guess you mean self._store?
And the issue now is that the line buf = self._store[f"{level}/{y}.{x}.0"] generates error, for example: KeyError: '7/0.0.0'. And here are the keys in self._store: dict_keys(['.zgroup', '.zattrs', '0/.zarray', '1/.zarray', '2/.zarray', '3/.zarray', '4/.zarray', '5/.zarray', '6/.zarray', '7/.zarray', '8/.zarray', '9/.zarray', '10/.zarray']).

How should I access the array data with f"{level}/{y}.{x}.0" in this case?

from vizarr.

manzt commented on June 13, 2024

You wrote store[...] in your previous code, I guess you mean self._store?

Oh that's my mistake. The code snippet I shared I just typed here as an idea and didn't test at all.

self._store is just a dict containing the zarr metadata. You should call the parent class (OpenSlideStore) __getitem__ method.

def __getitem__(key):
  # ....
  buf = super().__getitem__(f"{level}/{y}.{x}.0")

from vizarr.

oeway commented on June 13, 2024

@manzt Thank you! It's working now! Awesome!

For others who might be interested in this, here is the working version for me:

import numpy as np
from napari_lazy_openslide import OpenSlideStore
from zarr.util import json_dumps, json_loads

def _parse_chunk_path(path: str):
    """Returns x,y chunk coords and pyramid level from string key"""
    level, ckey = path.split("/")
    c, y, x  = map(int, ckey.split("."))
    return c, x, y, int(level)

class OpenSlideStoreWrapper(OpenSlideStore):

    def __getitem__(self, key):
        if key in self._store: # metadata store for OpenSlideStore
            # key is for metadata
            res = self._store[key]
        if key.endswith('.zarray'):
            meta = json_loads(res)
            meta["shape"] = [meta["shape"][2], meta["shape"][0], meta["shape"][1]]
            meta["chunks"] = [1, meta["chunks"][0], meta["chunks"][1]]
            return json_dumps(meta)
        elif key.endswith('.zgroup') or key.endswith('.zattrs'):
            # zgroup
            return res

        cidx, y, x, level = _parse_chunk_path(key)
        buf = super().__getitem__(f"{level}/{x}.{y}.0") # could cache this result
        arr = np.frombuffer(buf, dtype='uint8').reshape(self._tilesize, self._tilesize, -1)
        data = arr[:, :, cidx] # cou
        return data.tobytes()

How do I make it display in color mode, I guess that has to do with the "omero" key, right? What's the best way to pass that? Another minor issue is that, if I zoom in too much it becomes blank, can I limit the zoom level?

BTW, would you consider having such feature in napari_lazy_openslide so it can work out of the box?

from vizarr.

oeway commented on June 13, 2024

All works now, see a gif below:

The .as_ngff() method sounds awesome! Looking forward to it!

from vizarr.

View zarr loaded with napari_lazy_openslide about vizarr HOT 6 CLOSED

Comments (6)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent