Wrong window metadata in projected image?

[tcompa]

See also https://github.com/fractal-analytics-platform/fractal-deployments/issues/116#issuecomment-2812444135.

I have these two plates:

18M	./20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr
35M	./20200812-CardiomyocyteDifferentiation14-Cycle1.zarr

created with v1.5.0a2.

I ran

$ pip install "napari[all]" "napari-ome-zarr"

And then

$ napari --plugin napari-ome-zarr ./20200812-CardiomyocyteDifferentiation14-Cycle1.zarr/
$ napari --plugin napari-ome-zarr ./20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr/

3D image

2D image

Here are the napari --verbose outputs.

3D image

napari --plugin napari-ome-zarr ./20200812-CardiomyocyteDifferentiation14-Cycle1.zarr/ --verbose
12:35:50 : INFO : MainThread : root_attr: plate
12:35:50 : INFO : MainThread : plate_data: {'acquisitions': [{'id': 0, 'name': '20200812-CardiomyocyteDifferentiation14-Cycle1'}], 'columns': [{'name': '03'}], 'rows': [{'name': 'B'}], 'version': '0.4', 'wells': [{'columnIndex': 0, 'path': 'B/03', 'rowIndex': 0}]}
12:35:50 : INFO : MainThread : root_attr: well
12:35:50 : INFO : MainThread : well_data: {'images': [{'path': '0'}], 'version': '0.4'}
12:35:50 : INFO : MainThread : root_attr: multiscales
12:35:50 : INFO : MainThread : root_attr: omero
12:35:50 : INFO : MainThread : datasets [{'coordinateTransformations': [{'scale': [1, 1.0, 0.1625, 0.1625], 'type': 'scale'}], 'path': '0'}, {'coordinateTransformations': [{'scale': [1, 1.0, 0.325, 0.325], 'type': 'scale'}], 'path': '1'}, {'coordinateTransformations': [{'scale': [1, 1.0, 0.65, 0.65], 'type': 'scale'}], 'path': '2'}, {'coordinateTransformations': [{'scale': [1, 1.0, 1.3, 1.3], 'type': 'scale'}], 'path': '3'}, {'coordinateTransformations': [{'scale': [1, 1.0, 2.6, 2.6], 'type': 'scale'}], 'path': '4'}]
12:35:50 : INFO : MainThread : resolution: 0
12:35:50 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 2, 2160, 5120)
12:35:50 : INFO : MainThread :  - chunks =  ['1', '2', '2160', '2560']
12:35:50 : INFO : MainThread :  - dtype = uint16
12:35:50 : INFO : MainThread : resolution: 1
12:35:50 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 2, 1080, 2560)
12:35:50 : INFO : MainThread :  - chunks =  ['1', '2', '1080', '2560']
12:35:50 : INFO : MainThread :  - dtype = uint16
12:35:50 : INFO : MainThread : resolution: 2
12:35:50 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 2, 540, 1280)
12:35:50 : INFO : MainThread :  - chunks =  ['1', '2', '540', '1280']
12:35:50 : INFO : MainThread :  - dtype = uint16
12:35:50 : INFO : MainThread : resolution: 3
12:35:50 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 2, 270, 640)
12:35:50 : INFO : MainThread :  - chunks =  ['1', '2', '270', '640']
12:35:50 : INFO : MainThread :  - dtype = uint16
12:35:50 : INFO : MainThread : resolution: 4
12:35:50 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 2, 135, 320)
12:35:50 : INFO : MainThread :  - chunks =  ['1', '2', '135', '320']
12:35:50 : INFO : MainThread :  - dtype = uint16
12:35:50 : INFO : MainThread : root_attr: multiscales
12:35:50 : INFO : MainThread : root_attr: omero

2D image:

12:36:01 : INFO : MainThread : root_attr: plate
12:36:01 : INFO : MainThread : plate_data: {'acquisitions': [{'id': 0, 'name': '20200812-CardiomyocyteDifferentiation14-Cycle1'}], 'columns': [{'name': '03'}], 'rows': [{'name': 'B'}], 'version': '0.4', 'wells': [{'columnIndex': 0, 'path': 'B/03', 'rowIndex': 0}]}
12:36:01 : INFO : MainThread : root_attr: well
12:36:01 : INFO : MainThread : well_data: {'images': [{'path': '0'}], 'version': '0.4'}
12:36:01 : INFO : MainThread : root_attr: multiscales
12:36:01 : INFO : MainThread : root_attr: omero
12:36:01 : INFO : MainThread : datasets [{'coordinateTransformations': [{'scale': [1.0, 1.0, 0.1625, 0.1625], 'type': 'scale'}], 'path': '0'}, {'coordinateTransformations': [{'scale': [1.0, 1.0, 0.325, 0.325], 'type': 'scale'}], 'path': '1'}, {'coordinateTransformations': [{'scale': [1.0, 1.0, 0.65, 0.65], 'type': 'scale'}], 'path': '2'}, {'coordinateTransformations': [{'scale': [1.0, 1.0, 1.3, 1.3], 'type': 'scale'}], 'path': '3'}, {'coordinateTransformations': [{'scale': [1.0, 1.0, 2.6, 2.6], 'type': 'scale'}], 'path': '4'}]
12:36:01 : INFO : MainThread : resolution: 0
12:36:01 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 1, 2160, 5120)
12:36:01 : INFO : MainThread :  - chunks =  ['1', '1', '2160', '2560']
12:36:01 : INFO : MainThread :  - dtype = uint16
12:36:01 : INFO : MainThread : resolution: 1
12:36:01 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 1, 1080, 2560)
12:36:01 : INFO : MainThread :  - chunks =  ['1', '1', '1080', '2560']
12:36:01 : INFO : MainThread :  - dtype = uint16
12:36:01 : INFO : MainThread : resolution: 2
12:36:01 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 1, 540, 1280)
12:36:01 : INFO : MainThread :  - chunks =  ['1', '1', '540', '1280']
12:36:01 : INFO : MainThread :  - dtype = uint16
12:36:01 : INFO : MainThread : resolution: 3
12:36:01 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 1, 270, 640)
12:36:01 : INFO : MainThread :  - chunks =  ['1', '1', '270', '640']
12:36:01 : INFO : MainThread :  - dtype = uint16
12:36:01 : INFO : MainThread : resolution: 4
12:36:01 : INFO : MainThread :  - shape ('c', 'z', 'y', 'x') = (1, 1, 135, 320)
12:36:01 : INFO : MainThread :  - chunks =  ['1', '1', '135', '320']
12:36:01 : INFO : MainThread :  - dtype = uint16
12:36:01 : INFO : MainThread : root_attr: multiscales
12:36:01 : INFO : MainThread : root_attr: omero

And here is the diff of the image zattrs:

30c30
<                                 1,
---
>                                 1.0,
44c44
<                                 1,
---
>                                 1.0,
58c58
<                                 1,
---
>                                 1.0,
72c72
<                                 1,
---
>                                 1.0,
86c86
<                                 1,
---
>                                 1.0,
96a97
>             "name": "MIP",
108,111c109,112
<                     "end": 800,
<                     "max": 65535,
<                     "min": 0,
<                     "start": 110
---
>                     "end": 65535.0,
>                     "max": 65535.0,
>                     "min": 0.0,
>                     "start": 0.0
114,117c115
<         ],
<         "id": 1,
<         "name": "TBD",
<         "version": "0.4"
---
>         ]

[tcompa]

If I manually edit the 2D-image metadata, and use the same window start/end/min/max as the 3D one, then I see the image correctly.

[tcompa]

Possibly relevant: Is it expected that in https://github.com/fractal-analytics-platform/ngio/blob/6bf7c0e54b5c4d40a0025075c50389879d174a03/src/ngio/images/image.py#L478 we are not deriving the window property?

[jluethi]

Not deriving the window property sounds like a bug to me. I think the 2 valid options would be to derive the window property or to recalculate based on percentiles. I'd favor the derive approach.

Setting it to full range will make things appear black on our typical underexposed microscopy images (even though the data in there likely is still correct)

[lorenzocerrone]

The bug is going to be resolved with an upstream ngio fix.
The fix is already here fractal-analytics-platform/ngio#61 and will be realesed by Tuesday

[tcompa]

closed with #943