cell-data-loader 0.0.33__tar.gz → 0.0.35__tar.gz

{cell_data_loader-0.0.33 → cell_data_loader-0.0.35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cell_data_loader
-Version: 0.0.33
+Version: 0.0.35
 Summary: Converts general images of cells into formats and labels for deep learning pipelines
 Project-URL: Homepage, https://github.com/mleming/CellDataLoader
 Author-email: Matt Leming <mleming@mgh.harvard.edu>

{cell_data_loader-0.0.33 → cell_data_loader-0.0.35}/pyproject.toml

@@ -2,7 +2,8 @@
 
 name = "cell_data_loader"
 
-version = "0.0.33"
+version = "0.0.35"
+
 
 dependencies = [
   "numpy>=1.22.0",

{cell_data_loader-0.0.33 → cell_data_loader-0.0.35}/src/cell_data_loader/__init__.py

@@ -1,3 +1,3 @@
-__version__ = "0.0.33"
+__version__ = "0.0.35"
 
 from .cell_data_loader import *

{cell_data_loader-0.0.33 → cell_data_loader-0.0.35}/src/cell_data_loader/cell_data_loader.py

@@ -15,7 +15,7 @@ import torch,torchvision
 #torchvision.disable_beta_transforms_warning()
 from .util import *
 import torchvision.transforms as transforms
-
+import gc
 #import openslide
 #import czifile
 try:
@@ -31,8 +31,10 @@ class ImageLabelObject(BaseDataset):
 			dtype="torch",
 			gpu_ids="",
 			dim=(64,64),
-			n_channels = None):
+			n_channels = None,
+			filename_label = None):
 		self.filename = filename
+		self.filename_label = filename_label
 		self.image = None
 		self.label = None
 		self.boxlabel = None
@@ -53,7 +55,12 @@ class ImageLabelObject(BaseDataset):
 			return "czi"
 		else:
 			raise Exception("Image type unsupported: %s" % ext)
-	def get_image(self):
+	def get_image(self,read_filename_label=False):
+		if read_filename_label:
+			temp = self.image
+			tempf = self.filename
+			self.image = self.image2
+			self.filename = self.filename_label
 		if self.image is None:
 			if self.im_type() == "regular":
 				self.image = cv2.imread(self.filename)
@@ -99,7 +106,9 @@ class ImageLabelObject(BaseDataset):
 			if len(self.image.shape) != 3:
 				self.image = np.expand_dims(self.image,axis=2)
 				assert(len(self.image.shape) == 3)
-			if self.n_channels is not None and \
+			if read_filename_label:
+				self.filename_label_channels = self.image.shape[2]
+			elif self.n_channels is not None and \
 					self.image.shape[2] != self.n_channels:
 				self.image = resample(self.image,
 					self.n_channels,axis=2)
@@ -110,6 +119,19 @@ class ImageLabelObject(BaseDataset):
 				assert(len(self.image.shape) == 3)
 			else:
 				raise Exception("Unimplemented dtype: %s" % self.dtype)
+
+			if read_filename_label:
+				if self.dtype == "torch":
+					assert(self.image.size()[0] == temp.size()[0])
+					assert(self.image.size()[1] == temp.size()[1])
+					self.image = torch.cat((temp,self.image),dim=2)
+				elif self.dtype == "numpy":
+					assert(self.image.shape[0] == temp.shape[0])
+					assert(self.image.shape[1] == temp.shape[1])
+					self.image = np.concatenate((temp,self.image),axis=2)
+				self.filename = tempf
+			elif self.filename_label is not None:
+				self.get_image(read_filename_label = True)
 		return self.image
 	def get_cell_box_label(self):
 		if self.boxlabel is None:
@@ -165,14 +187,12 @@ class ImageLabelObject(BaseDataset):
 	def __getitem__(self,index):
 		if self.mode == "whole":
 			im = self.get_image()
-			return im
 		elif self.mode == "sliced":
 			x_dim,y_dim = self.get_scaled_dims()
 			x = index % (x_dim)
 			y = (index // (x_dim)) % (y_dim)
-			imslice = self.get_image()[x * self.dim[0]:(x+1)*self.dim[0],
+			im = self.get_image()[x * self.dim[0]:(x+1)*self.dim[0],
 				y * self.dim[1]:(y+1)*self.dim[1],...]
-			return imslice
 		elif self.mode == "cell":
 			if len(self.get_cell_box_label()) == 0:
 				#print("len self: %d" % len(self))
@@ -182,14 +202,17 @@ class ImageLabelObject(BaseDataset):
 				#return -1
 			index = index % len(self.get_cell_box_label())
 			x,y,l,w = self.get_cell_box_label()[index]
-			return slice_and_augment(self.get_image(),x,y,l,w,
+			im = slice_and_augment(self.get_image(),x,y,l,w,
 				out_size=self.dim)
-			raise Exception("Unimplemented")
 		else:
 			raise Exception("Invalid mode: %s" % self.mode)
+		return im
 	def clear(self):
 		del self.image
+		del self.boxlabel
+		gc.collect()
 		self.image = None
+		self.boxlabel = None
 
 class CellDataloader():#BaseDataset):
 	def __init__(self,
@@ -212,7 +235,8 @@ class CellDataloader():#BaseDataset):
 		normalize=True,
 		split = None,
 		return_filenames = False,
-		sample_output_folder = None):
+		sample_output_folder = None,
+		save_ram = False):
 		
 		self.verbose = verbose
 		self.label_balance = label_balance
@@ -232,15 +256,16 @@ class CellDataloader():#BaseDataset):
 		self.normalize = normalize
 		self.return_filenames = return_filenames
 		self.sample_output_folder = sample_output_folder
+		self.save_ram = save_ram
 		
 		if self.augment_image and self.dtype == "torch":
 			self.augment = transforms.Compose([
 				transforms.RandomHorizontalFlip(0.5),
-				transforms.RandomVerticalFlip(0.5),
-				transforms.GaussianBlur(5),
-				transforms.ColorJitter(brightness=(0.5,1.5),
-					contrast=(1), saturation=(0.5,1.5), hue=(-0.1,0.1)),
-				transforms.RandomResizedCrop(self.dim, antialias=True)])#,
+				transforms.RandomVerticalFlip(0.5)])
+				#transforms.GaussianBlur(5),
+				#transforms.ColorJitter(brightness=0.1,
+				#	contrast=0.1, saturation=0.1, hue=0.1),
+				#transforms.RandomResizedCrop(self.dim, antialias=True)])#,
 				#transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5))])
 			self.augment2 = transforms.Compose([
 				transforms.ElasticTransform()])
@@ -513,7 +538,8 @@ class CellDataloader():#BaseDataset):
 		self.im_index += 1
 		while self.im_index >= len(self.image_objects[self.index]) or \
 				(len(self.image_objects[self.index]) == 0):
-			self.image_objects[self.index].clear()
+			if self.save_ram:
+				self.image_objects[self.index].clear()
 			self.im_index = 0
 			self.index += 1
 			if self.index >= len(self.image_objects): break
@@ -594,8 +620,12 @@ class CellDataloader():#BaseDataset):
 				out_file_name = os.path.join(self.sample_output_folder,
 					"%.8d" % self.imcount)
 				if self.return_labels():
-					out_file_name = "%s_%d" % (out_file_name,
+					if self.dtype == "numpy":
+						out_file_name = "%s_%d" % (out_file_name,
 										int(np.argmax(self.label_batch[i])))
+					elif self.dtype == "torch":
+						out_file_name = "%s_%d" % (out_file_name,
+									int(torch.argmax(self.label_batch[i])))
 				if self.return_filenames:
 					out_file_name = "%s_%s" % (out_file_name,
 						os.path.basename(os.path.splitext(fnames[i])[0]))

{cell_data_loader-0.0.33 → cell_data_loader-0.0.35}/src/cell_data_loader/util.py

@@ -415,6 +415,8 @@ def merge_predicted_and_annotated(annotated_boxes,predicted_boxes,min_dice=0.5):
 	return np.concatenate((annotated_boxes,intersected_boxes),axis=0)
 
 def get_average_mask_size(mask):
+	m = mask.max()
+	if m == 0: return 0
 	return np.sum(mask.flatten() > 0) / mask.max()
 
 def renumber_masks(mask):