Download Data¶

In this notebook, we will download data and convert it to a zarr dataset.
This tutorial was written by Henry Westmacott and Manan Lalit.

For demonstration, we will use a subset of images of Fluo-N2DL-HeLa available on the Cell Tracking Challenge webpage.

Firstly, the tif raw images are downloaded to a directory indicated by data_dir.

In [1]:

from pathlib import Path

from pathlib import Path

In [2]:

import numpy as np
import tifffile
import zarr
from cellulus.utils.misc import extract_data
from csbdeep.utils import normalize
from tqdm import tqdm

import numpy as np import tifffile import zarr from cellulus.utils.misc import extract_data from csbdeep.utils import normalize from tqdm import tqdm

In [3]:

name = "2d-data-demo"
data_dir = "./data"

extract_data(
    zip_url="https://github.com/funkelab/cellulus/releases/download/v0.0.1-tag/2d-data-demo.zip",
    data_dir=data_dir,
    project_name=name,
)

name = "2d-data-demo" data_dir = "./data" extract_data( zip_url="https://github.com/funkelab/cellulus/releases/download/v0.0.1-tag/2d-data-demo.zip", data_dir=data_dir, project_name=name, )

Created new directory ./data

Downloaded and unzipped data to the location ./data

Next, a channel dimension is added to these images and they are appended in a list.

In [4]:

container_path = zarr.open(name + ".zarr")
dataset_name = "train/raw"
image_filenames = sorted((Path(data_dir) / name / "images").glob("*.tif"))
print(f"Number of raw images is {len(image_filenames)}")
image_list = []

for i in tqdm(range(len(image_filenames))):
    im = normalize(
        tifffile.imread(image_filenames[i]).astype(np.float32), 1, 99.8, axis=(0, 1)
    )
    image_list.append(im[np.newaxis, ...])

image_list = np.asarray(image_list)

container_path = zarr.open(name + ".zarr") dataset_name = "train/raw" image_filenames = sorted((Path(data_dir) / name / "images").glob("*.tif")) print(f"Number of raw images is {len(image_filenames)}") image_list = [] for i in tqdm(range(len(image_filenames))): im = normalize( tifffile.imread(image_filenames[i]).astype(np.float32), 1, 99.8, axis=(0, 1) ) image_list.append(im[np.newaxis, ...]) image_list = np.asarray(image_list)

Number of raw images is 11

  0%|                                                                                                                                                              | 0/11 [00:00<?, ?it/s]

 55%|█████████████████████████████████████████████████████████████████████████████████▊                                                                    | 6/11 [00:00<00:00, 54.47it/s]

100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 11/11 [00:00<00:00, 57.27it/s]

Lastly, the zarr dataset is populated, the axis names and resolution is specified.

In [5]:

container_path[dataset_name] = image_list
container_path[dataset_name].attrs["resolution"] = (1, 1)
container_path[dataset_name].attrs["axis_names"] = ("s", "c", "y", "x")

container_path[dataset_name] = image_list container_path[dataset_name].attrs["resolution"] = (1, 1) container_path[dataset_name].attrs["axis_names"] = ("s", "c", "y", "x")