# coding: utf-8 # # NAME: CellPose Segmenter # FILE: CellPose_Segmenter_RevA # REVISION : 1.4.1 - 2022-05-14 # AUTHOR : Maurizio Abbate / Mariia Burdyniuk # Giovanni Cardone (Biochem-MPG) # Copyright(c) 2021 arivis AG, Germany. All Rights Reserved. # # Permission is granted to use, modify and distribute this code, # as long as this copyright notice remains part of the code. # PURPOSE : # Tested for V4d Release : 3.6 # NOTE: Tiling has been added to face huge dataset # ------------------------------ External Package Import ---------------------- import arivis_parameter, arivis_operation import arivis_core as core import numpy as np from cellpose import models # import cv2 # from arivis_objects import Polygon # from arivis_core import Point2D # import arivis_objects as objects @arivis_parameter.add_arivis_parameters(Model_Name = 'cyto2', \ Input_channel = 1,\ Second_channel = 0,\ Diameter_in_um = 3.,\ Flow_threshold = arivis_parameter.param_range(0.4,0.01,1), \ Cellprob_threshold = arivis_parameter.param_range(0,-6,6),\ Fast_3D=True, Custom_model = False, Custom_model_path = 'C:/' ,\ Tile_Size = arivis_parameter.param_range(4096,1024,65536)) @arivis_parameter.add_param_description(Model_Name = 'Choose from {cyto, cyto2, nuclei}',\ Input_channel = "Main prediction channel",\ Second_channel= 'Additional channel for the prediction (e.g. nuclei for Cyto2 model)',\ Diameter_in_um = 'Estimated diameter in microns (0:auto estimation)',\ Flow_threshold = 'Consistency with model (0...1)',\ Cellprob_threshold = 'Mask inclusion threshold (-6...+6)',\ Fast_3D='Analyse slices separately',\ Custom_model = 'Do you have your own model?',\ Custom_model_path = 'Model path including hte file name' ,\ Tile_Size = "Max tile size (XY - pizels)" ) def main(Model_Name, Input_channel, Second_channel, Diameter_in_um, Flow_threshold, Cellprob_threshold, Fast_3D, Custom_model, Custom_model_path,Tile_Size): # --------------------------------------------------------------------------- # make tensorflow output silent: # --------------------------------------------------------------------------- import os os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' # or any {'0', '1', '2'} # --------------------------------------------------------------------------- # suppress syntax warning from tensorflow # --------------------------------------------------------------------------- import warnings warnings.filterwarnings("ignore", category=SyntaxWarning) np.random.seed(6) # configure model if not Custom_model: model = models.Cellpose(gpu=True, model_type=Model_Name) else: model = models.CellposeModel(gpu=True, pretrained_model=Custom_model_path, net_avg=False) # --------------------------------------------------------------------------- #check if the input is two channel or one: # --------------------------------------------------------------------------- if Second_channel >= 1: channels_CP = [1,2] else: channels_CP = [1,0] Input_channel -= 1 # change to zero-based indexing Second_channel -= 1 context = arivis_operation.Operation.get_context() input_data = context.get_input() output_data = context.get_output() input_bounds = input_data.get_bounds() #bounding box 4D pixel_type = input_data.get_pixel_type() pixel_size = input_data.get_pixel_size() # --------------------------------------------------------------------------- # check the diameter: # --------------------------------------------------------------------------- Diameter_in_pixels = int (Diameter_in_um / pixel_size.x) if Diameter_in_um == 0 or Diameter_in_pixels == 0: diameter = None else: diameter = Diameter_in_pixels number_of_channels = input_data.get_channel_count() if(number_of_channels < 1): print("Error: this script needs at least one channel. End of script.") number_of_timepoints = (input_bounds.t2 - input_bounds.t1 + 1) number_of_planes = (input_bounds.z2 - input_bounds.z1 + 1) # --------------------------------------------------------------------------- totalWork = number_of_timepoints workDone = 0 # --------------------------------------------------------------------------- tile = core.Bounds3D() tile.x1 = input_bounds.x1 tile.x2 = input_bounds.x2 tile.y1 = input_bounds.y1 tile.y2 = input_bounds.y2 tile.z1 = input_bounds.z1 tile.z2 = input_bounds.z2 # --------------------------------------------------------------------------- # check the anisotropy: # --------------------------------------------------------------------------- if number_of_planes > 1: process_3D = not Fast_3D stitch_thr = 0.2 anisotropy = pixel_size.z/pixel_size.x # assuming same x and y else: process_3D = False stitch_thr = 0 anisotropy = 1.0 # --------------------------------------------------------------------------- # get tiles # --------------------------------------------------------------------------- tiles=[] tiles.append(tile) if Tile_Size < max(tile.x2-tile.x1+1,tile.y2-tile.y1+1): tiles=[] tiles = core.get_tiles(input_bounds, Tile_Size, overlap=0) # debug numero = len(tiles) # --------------------------------------------------------------------------- # run tiles # --------------------------------------------------------------------------- for timepoint in range(input_bounds.t1, input_bounds.t2 + 1): # ----------------------------------------------------------------------- for tile1 in tiles: # ------------------------------------------------------------------- if Second_channel >= 1: buffer = np.empty((tile1.z2 - tile1.z1 + 1, tile1.y2 - tile1.y1 + 1, tile1.x2 - tile1.x1 + 1, 2), dtype=pixel_type) buffer[:,:,:,0] = input_data.read_imagedata(region=tile1, channel=Input_channel, timepoint=timepoint) buffer[:,:,:,1] = input_data.read_imagedata(region=tile1, channel=Second_channel, timepoint=timepoint) else: buffer = np.empty((tile1.z2 - tile1.z1 + 1, tile1.y2 - tile1.y1 + 1, tile1.x2 - tile1.x1 + 1), dtype=pixel_type) buffer = input_data.read_imagedata(region=tile1, channel=Input_channel, timepoint=timepoint) # ------------------------------------------------------------------- # model.Eval # ------------------------------------------------------------------- if not Custom_model: masks, flows, styles, diams = model.eval(buffer, diameter=diameter, channels=channels_CP, flow_threshold=Flow_threshold, cellprob_threshold=Cellprob_threshold, stitch_threshold=stitch_thr, anisotropy=anisotropy, do_3D=process_3D) else: masks, flows, styles = model.eval(buffer, diameter=diameter, channels=channels_CP, flow_threshold=Flow_threshold, cellprob_threshold=Cellprob_threshold, stitch_threshold=stitch_thr, anisotropy=anisotropy, do_3D=process_3D) # ------------------------------------------------------------------- # prepare segmentBuilder # ------------------------------------------------------------------- segmentBuilder = arivis_operation.SegmentBuilder.segment_builder(input_data, output_data) with segmentBuilder.at_timepoint(timepoint): print('tile_bounds is '+ str(tile1.__dict__)) segmentBuilder.create_segments(masks, tile1) # --------------------------------------------------------------- workDone = workDone + 1 context.notify_progress(workDone * 100 / totalWork)