Run Cellpose with appose

Now let’s do something more exciting - run Cellpose segmentation using Appose! This shows how to send real data (an image) to another environment and get meaningful results back.

Important: Set your notebook kernel to ‘Python (appose_napari_ai)’ - this environment doesn’t have Cellpose installed, which perfectly demonstrates calling Cellpose from a different environment!

Load Libraries and Check Local Environment

First, let’s import what we need. Notice we check if cellpose is available locally - it probably isn’t, and that’s okay! We’ll run it in another environment.

from skimage.io import imread
from matplotlib import pyplot as plt
import appose

try:
    import cellpose
except ImportError:
    print("Cellpose is not installed.")

Cellpose is not installed.

Load Our Test Image

Let’s load the same cell image we’ve been using. This will be sent to the Cellpose environment for segmentation.

image_path = r"..\..\data\SOTA_segmentation\cell_00068.bmp"
img = imread(image_path)
plt.imshow(img)

<matplotlib.image.AxesImage at 0x1d06f8a6810>

Write the Cellpose Code

We write Cellpose code as a string that will run in the other environment. Note how we get the result back through task.outputs.

# Create the execution string
execution_string = f'''
import numpy as np
from cellpose import models
import appose

model = models.CellposeModel(gpu=True)
    
masks, flows, styles = model.eval(
        image.ndarray(),
        #diameter=30,
        flow_threshold=0.5,
        niter=200
)

# Convert result to appose format for output

ndarr_mask = appose.NDArray(dtype=str(masks.dtype), shape=masks.shape)
ndarr_mask.ndarray()[:] = masks

task.outputs["mask"] = ndarr_mask
'''

Prepare the Image for Sending

We need to convert our image into Appose’s format so it can be sent to the other environment.

cellpose_env_path = r"..\..\pixi\microsam_cellposesam\.pixi\envs\default"

env = appose.base(cellpose_env_path).build()
ndarr_img = appose.NDArray(dtype=str(img.dtype), shape=img.shape)
ndarr_img.ndarray()[:] = img

# Always include image, optionally add additional inputs
inputs = {"image": ndarr_img}

Run Cellpose in the Other Environment

Now we send our image and code to the Cellpose environment, wait for it to finish, and get our segmentation mask back.

with env.python() as python:
    task = python.task(execution_string, inputs=inputs, queue="main")
    task.wait_for()

    if task.error:
        print(f"⚠️  Task error: {task.error}")

    print(task.outputs.keys())

    result = task.outputs.get("mask", None)

⚠️  Task error: Traceback (most recent call last):
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\appose\python_worker.py", line 145, in _run
    exec(compile(block, "<string>", mode="exec"), _globals, binding)
  File "<string>", line 8, in <module>
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\models.py", line 338, in eval
    masks = self._compute_masks(x.shape, dP, cellprob, flow_threshold=flow_threshold,
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\models.py", line 524, in _compute_masks
    outputs = dynamics.resize_and_compute_masks(
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\dynamics.py", line 610, in resize_and_compute_masks
    mask = compute_masks(dP, cellprob, niter=niter,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\dynamics.py", line 672, in compute_masks
    mask = remove_bad_flow_masks(mask, dP, threshold=flow_threshold,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\dynamics.py", line 443, in remove_bad_flow_masks
    merrors, _ = flow_error(masks, flows, device0)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\dynamics.py", line 300, in flow_error
    dP_masks = masks_to_flows_gpu(maski, device=device)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\dynamics.py", line 139, in masks_to_flows_gpu
    mu = _extend_centers_gpu(neighbors, meds_p, isneighbor, shape, n_iter=n_iter,
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\Users\bnort\work\ImageJ2022\tnia\notebooks-and-napari-widgets-for-dl\pixi\microsam_cellposesam\.pixi\envs\default\Lib\site-packages\cellpose\dynamics.py", line 45, in _extend_centers_gpu
    Tneigh *= isneighbor
RuntimeError: CUDA error: an illegal memory access was encountered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.


dict_keys([])

Compare Results

Let’s see how well our Appose-powered Cellpose did! We’ll compare it side-by-side with the ground truth.

# Load ground truth for comparison
label_path = r"..\..\data\SOTA_segmentation\cell_00068_label.tiff"
ground_truth = imread(label_path)

# Import visualization helper
from tnia.plotting.plt_helper import mask_overlay

# Create overlays
ground_truth_overlay = mask_overlay(img, ground_truth)
result_overlay = mask_overlay(img, result.ndarray())

# Create comparison plot
fig, axes = plt.subplots(1, 2, figsize=(12, 6))

axes[0].imshow(ground_truth_overlay)
axes[0].set_title('Ground Truth')
axes[0].axis('off')

axes[1].imshow(result_overlay)
axes[1].set_title('Appose Cellpose Result')
axes[1].axis('off')

plt.tight_layout()
plt.show()