rossbar
commented
1 year ago Without any more specific information, it's difficult to say definitively where the problem lies, but given the info above it almost certainly lies in the data and problem setup. Generally speaking, the issue tracker is for reporting issues with the library, not user code - I will go ahead and close this as there isn't anything actionable here.
If you end up with a minimal reproducible example of something that you believe stems from a bug in the library, please feel free to reopen.
ngreenwald
In the below code I am trying to segment a TIFF image converted to NPZ. The code runs without error but the segmentation output is not as desired. Screenshot of the final output attached for reference.
!/usr/bin/env python
coding: utf-8
In[1]:
import numpy as np import matplotlib.pyplot as plt import tifffile as tiff import os import pandas as pd
In[2]:
img = tiff.imread('A9D035F9-5A78-47A6-8C41-A1A35487CEF9_codex-ome copy.tif') img_array = np.array(img)
In[3]:
split_ratio = 0.5 # You can adjust this ratio based on your requirements split_index = int(len(img_array) * split_ratio) train_data1 = img_array[:split_index] test_data1 = img_array[split_index:]
In[4]:
Save the train and test data in separate NPZ files
np.savez('tissuenet_v1.0_train.npz', data=train_data1) np.savez('tissuenet_v1.0_test.npz', data=test_data1)
In[5]:
train_data = np.load('tissuenet_v1.0_train.npz') test_data = np.load('tissuenet_v1.0_test.npz')
In[6]:
print(train_data['data'].shape) print(test_data['data'].shape)
In[7]:
new_shape = (2601, 512, 512, 2) data_1 = train_data['data'] train_data_reshaped = data_1[:2601, :512, :512] train_data_reshaped = np.stack((train_data_reshaped, train_data_reshaped), axis=-1)
In[8]:
data_2 = test_data['data'] test_data_reshaped = data_2[:2601, :512, :512] test_data_reshaped = np.stack((test_data_reshaped, test_data_reshaped), axis=-1)
In[9]:
train_data_reshaped = np.load('train_data_reshaped.npz') test_data_reshaped = np.load('test_data_reshaped.npz')
In[10]:
train_data_reshaped = train_data_reshaped['data']
X_train = train_data_reshaped[:, :, :, :-1] # Features, all dimensions except the last one y_train = train_data_reshaped[:, :, :, :-1] # Labels, the last dimension
In[11]:
test_data_reshaped = test_data_reshaped['data']
X_test = train_data_reshaped[:, :, :, :-1] # Features, all dimensions except the last one y_test = train_data_reshaped[:, :, :, :-1] # Labels, the last dimension
In[12]:
print(X_train.shape) print(y_train.shape)
In[13]:
print(X_test.shape) print(y_test.shape)
In[14]:
from skimage.transform import resize
Resize X_train and y_train to the desired shape
desired_shape = (256, 256, 1) X_train_resized = np.zeros((X_train.shape[0],) + desired_shape) y_train_resized = np.zeros((y_train.shape[0],) + desired_shape)
for i in range(X_train.shape[0]): X_train_resized[i] = resize(X_train[i, :, :, 0], desired_shape, anti_aliasing=True, mode='reflect') y_train_resized[i] = resize(y_train[i, :, :, 0], desired_shape, anti_aliasing=True, mode='reflect')
Now X_train_resized and y_train_resized should have the shape (256, 256, 1)
In[15]:
from skimage.transform import resize
Resize X_train and y_train to the desired shape
desired_shape_test = (256, 256, 1) X_test_resized = np.zeros((X_test.shape[0],) + desired_shape_test) y_test_resized = np.zeros((y_test.shape[0],) + desired_shape_test)
for i in range(X_train.shape[0]): X_test_resized[i] = resize(X_test[i, :, :, 0], desired_shape_test, anti_aliasing=True, mode='reflect') y_test_resized[i] = resize(y_test[i, :, :, 0], desired_shape_test, anti_aliasing=True, mode='reflect')
Now X_test_resized and y_test_resized should have the shape (256, 256, 1)
In[16]:
test_size = 0.5 # % of data saved as test seed = 0 # seed for random train-test split
In[17]:
from deepcell.utils.data_utils import reshape_matrix
size = 256
X_train, y_train = reshape_matrix(X_train_resized, y_train_resized, reshape_size=size) print('X.shape: {}\ny.shape: {}'.format(X_train.shape, y_train.shape))
In[18]:
from deepcell.utils.data_utils import reshape_matrix
size = 256
X_test, y_test = reshape_matrix(X_test_resized, y_test_resized, reshape_size=size) print('X.shape: {}\ny.shape: {}'.format(X_test.shape, y_test.shape))
In[19]:
import os os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
from deepcell.model_zoo.panopticnet import PanopticNet
classes = { 'inner_distance': 1, # inner distance 'outer_distance': 1, # outer distance 'fgbg': 2, # foreground/background separation }
model = PanopticNet( backbone='resnet50', input_shape=X_train_resized.shape[1:], norm_method='std', num_semantic_classes=classes)
In[30]:
from tensorflow.keras.optimizers import SGD, Adam from deepcell.utils.train_utils import rate_scheduler
model_name = 'watershed_centroid_nuclear_general_std'
n_epoch = 5 # Number of training epochs test_size = .20 # % of data saved as test norm_method = 'whole_image' # data normalization
lr = 1e-5 optimizer = Adam(learning_rate=lr, clipnorm=0.001) lr_sched = rate_scheduler(lr=lr, decay=0.99)
batch_size = 1
min_objects = 1 # throw out images with fewer than this many objects
In[31]:
from deepcell import image_generators from deepcell.utils import train_utils
transforms = list(classes.keys()) transforms_kwargs = {'outer-distance': {'erosion_width': 0}}
use augmentation for training but not validation
datagen = image_generators.SemanticDataGenerator( rotation_range=180, shear_range=0, zoom_range=(0.75, 1.25), horizontal_flip=True, vertical_flip=True)
datagen_val = image_generators.SemanticDataGenerator( rotation_range=0, shear_range=0, zoom_range=0, horizontal_flip=0, vertical_flip=0)
train_data = datagen.flow( {'X': X_train, 'y': y_train}, seed=seed, transforms=transforms, transforms_kwargs=transforms_kwargs, min_objects=min_objects, batch_size=batch_size)
val_data = datagen_val.flow( {'X': X_test, 'y': y_test}, seed=seed, transforms=transforms, transforms_kwargs=transforms_kwargs, min_objects=min_objects, batch_size=batch_size)
In[32]:
from deepcell.model_zoo.panopticnet import PanopticNet
classes = { 'inner_distance': 1, 'outer_distance': 1, }
prediction_model = PanopticNet( backbone='resnet50', input_shape=X_train.shape[1:], norm_method='std', num_semantic_heads=2, num_semantic_classes=classes, location=True, # should always be true include_top=True)
In[33]:
from timeit import default_timer
start = default_timer() test_images = prediction_model.predict(X_test) watershed_time = default_timer() - start
print('Watershed segmentation of shape', test_images[0].shape, 'in', watershed_time, 'seconds.')
In[34]:
import time
from matplotlib import pyplot as plt import numpy as np
from skimage.feature import peak_local_max
from deepcell_toolbox.deep_watershed import deep_watershed
index = 1000
fig, axes = plt.subplots(1, 4, figsize=(20, 20))
masks = deep_watershed( test_images, min_distance=10, detection_threshold=0.1, distance_threshold=0.01, exclude_border=False, small_objects_threshold=0)
calculated in the postprocessing above, but useful for visualizing
inner_distance = test_images[0] outer_distance = test_images[1]
coords = peak_local_max( inner_distance[index], min_distance=10, threshold_abs=0.1, exclude_border=False)
raw image with centroid
axes[0].imshow(X_test[index, ..., 0]) axes[0].scatter(coords[..., 1], coords[..., 0], color='r', marker='.', s=10)
axes[1].imshow(inner_distance[index, ..., 0], cmap='jet') axes[2].imshow(outer_distance[index,
..., 0], cmap='jet') axes[3].imshow(masks[index, ...], cmap='jet')
plt.show()