Error with default `spectral_bins_per_emission_channel=1` when using COSEM spectra

[Lil-Conch]

Hi, tlambert!
I’m getting the following error when running a simulation with COSEM fluorophores using the default settings:

ValueError: dimensions () must have the same length as the number of data dimensions, ndim=1

It happens inside _simple_psf.py at

bins = np.linspace(w_min, w_max, num_bins + 1)

when spectral_bins_per_emission_channel = 1 (the default).

From the stack trace it looks like w_min / w_max are xarray.DataArray scalars (from COSEM spectra), so np.linspace gets intercepted by xarray’s __array_wrap__, which then fails on dimension handling.

My guess is that this path was written assuming plain numpy floats, but COSEM uses xarray for spectra. Possibly w_min / w_max just need to be converted to float before calling linspace, or num_bins=1 should directly use the spectral centroid instead of [w_min, w_max].

Does that sound right?

Thanks!

[tlambert03]

hi @Lil-Conch, it would be much easier to help if I could see the code you're running, not just the exception. Can you paste the contents of your cosem_jrc_159-4.py file so I can try to reproduce it?

[Lil-Conch]

hi @tlambert03. Here's my simulation script.

from microsim import schema as ms
from microsim.schema.optical_config import lib
import numpy as np
import tifffile  

shape = (512, 2048, 2048)
scale = 0.016
ph_au = 1.
downscale=(1, 1, 1)

###################################################
green_cfg = ms.OpticalConfig.from_fpbase(
    microscope_id="wKqWbg",  
    config_name="Widefield Green",
)
red_cfg = ms.OpticalConfig.from_fpbase(
    microscope_id="wKqWbg",
    config_name="Widefield Red",
)
farred_cfg = ms.OpticalConfig.from_fpbase(
    microscope_id="wKqWbg",
    config_name="Widefield Far-Red",
)
blue_cfg = ms.OpticalConfig.from_fpbase(
    microscope_id="wKqWbg",
    config_name="Widefield Blue",
)

###################################################
dataset = "jrc_macrophage-2"
# dataset = ""
sample = [
    # Blue 通道（mTagBFP2）——核膜
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="ne-mem_pred"),
    #     fluorophore="mTagBFP2",
    # ),

    # Green 通道（EGFP 系）——内质网、内体、脂滴
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="er-mem_pred"),
    #     fluorophore="EGFP",
    # ),
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="endo-mem_pred"),
    #     fluorophore="mNeonGreen",
    # ),
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="ld-mem_pred"),
    #     fluorophore="Venus",
    # ),

    # Red 通道（mCherry / mScarlet 系）——线粒体、高尔基、小囊泡
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="mito-mem_pred"),
    #     fluorophore="mCherry",
    # ),
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="golgi-mem_pred"),
    #     fluorophore="mScarlet",
    # ),
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="vesicle-mem_pred"),
    #     fluorophore="TagRFP-T",
    # ),

    # Far-Red 通道（Alexa Fluor 647）——溶酶体
    # ms.FluorophoreDistribution(
    #     distribution=ms.CosemLabel(dataset=dataset, label="lyso-mem_pred"),
    #     fluorophore="Alexa Fluor 647",
    # ),
    ms.FluorophoreDistribution(
        distribution=ms.CosemLabel(dataset=dataset, label="mt-in_pred"),
        fluorophore="Alexa Fluor 647",
    ),
]


sim = ms.Simulation(
    # note: this is a rather coarse simulation, but it's fast
    # scale should be a one of .004 * 2^n, where n is an integer from 0 to 4
    # space basically determines the field of view.
    truth_space=ms.ShapeScaleSpace(
        shape=shape,
        scale=scale,
    ),
    output_space={"downscale": downscale},
    sample=sample,
    channels=[farred_cfg],
    # modality=ms.Widefield(),
    modality=ms.Confocal(pinhole_au=ph_au),
    output_path="au02.tiff",
)
result = sim.run()
# print("finish simulation")
# optical_image = sim.optical_image()
# print("converted to optical image")
digital = sim.digital_image()

tifffile.imwrite(f"{dataset}_farred.tif", np.asarray(digital))     

I'm trying to simulate the microtubules from jrc_macrophage-2 dataset. I've successfully run the script like a month before but encounter the error today. Currently I've replaced the function bin_spectrum in _simple_psf.py by the following content

w_min_ = float(w_min.values) if hasattr(w_min, "values") else float(w_min)
w_max_ = float(w_max.values) if hasattr(w_max, "values") else float(w_max)
bins = np.linspace(w_min_, w_max_, num_bins + 1)

where the original content is

bins = np.linspace(w_min, w_max, num_bins + 1)

I've also printed the w_min and w_max to validate the xarray structure:
w_min: <xarray.DataArray 'w' ()> Size: 8B np.float64(658.0) Coordinates: f object 8B Alexa Fluor 647 c object 8B Widefield Far-Red Attributes: units: nm, w_max: <xarray.DataArray 'w' ()> Size: 8B np.float64(736.0) Coordinates: f object 8B Alexa Fluor 647 c object 8B Widefield Far-Red Attributes: units: nm
The script successfully ran after the modification and the structure seems right. However I'm still not sure if it's my script's problem and whether the modification is proper. Thanks again for your help!

[tlambert03]

when running your script, I didn't get an error...
I did notice, however, that my output was different it used far red, and your exception was with widefield blue.

[15:42:46:243134] INFO     Ground truth generated in 27.16 seconds                                        simulation.py:153
[15:42:46:308068] INFO     Rendering Confocal channel Widefield Far-Red ---------------                   _simple_psf.py:55
[15:42:46:308751] INFO       Fluor: Alexa Fluor 647                                                       _simple_psf.py:61

however, when I modified the script to use the mTagBFP2 fluorophore and the blue_cfg, I still didn't get the error. So we probably need more info (might be environment differences). Can you please double check that the exact script you provided above is still giving you the error, and then paste the output of pip list so I can see the actual dependency versions you're working with?

[tlambert03]

(it's totally possible/probable that the fix you proposed is the right one, but I would really like to be able to reproduce the error and understand it before using that fix)