From 23b903d7852052e7e63a2e63b1a68ca0bd616353 Mon Sep 17 00:00:00 2001
From: Christopher Rhodes <christopher.rhodes@embl.de>
Date: Tue, 19 Dec 2023 15:46:55 +0100
Subject: [PATCH] Delegated export of patch products to ZMaskObjectTable class;
consolidated both batch runner and API-facing workflows into
infer_object_map_from_zstack(); consolidated parameterization of export
products into pydantic model
---
.../examples/export_patch_focus_metrics.py | 21 +-
extensions/chaeo/params.py | 24 +
extensions/chaeo/workflows.py | 510 +++++++++++-------
extensions/chaeo/zmask.py | 146 ++++-
extensions/ilastik/models.py | 62 ++-
extensions/ilastik/router.py | 6 +-
6 files changed, 527 insertions(+), 242 deletions(-)
create mode 100644 extensions/chaeo/params.py
diff --git a/extensions/chaeo/examples/export_patch_focus_metrics.py b/extensions/chaeo/examples/export_patch_focus_metrics.py
index 31258117..8e5fa332 100644
--- a/extensions/chaeo/examples/export_patch_focus_metrics.py
+++ b/extensions/chaeo/examples/export_patch_focus_metrics.py
@@ -9,7 +9,7 @@ from skimage.filters import gaussian, sobel
from extensions.ilastik.models import IlastikPixelClassifierModel
from extensions.chaeo.products import export_3d_patches_with_focus_metrics, export_patches_from_zstack
-from extensions.chaeo.zmask import build_zmask_from_object_mask
+from extensions.chaeo.zmask import ZMaskObjectTable
from model_server.accessors import generate_file_accessor, InMemoryDataAccessor, write_accessor_data_to_file
from model_server.workflows import Timer
@@ -51,20 +51,27 @@ def export_patch_focus_metrics_from_multichannel_zstack(
ti.click('threshold_pixel_mask')
# make zmask
- zmask, zmask_meta, df, interm = build_zmask_from_object_mask(
+ # zmask, zmask_meta, df, interm = build_zmask_from_object_mask(
+ # obmask.get_one_channel_data(pixel_class),
+ # stack.get_one_channel_data(zmask_channel),
+ # mask_type=mask_type,
+ # filters=zmask_filters,
+ # expand_box_by=zmask_expand_box_by,
+ # )
+ obj_table = ZMaskObjectTable(
obmask.get_one_channel_data(pixel_class),
stack.get_one_channel_data(zmask_channel),
mask_type=mask_type,
filters=zmask_filters,
expand_box_by=zmask_expand_box_by,
)
- zmask_acc = InMemoryDataAccessor(zmask)
+ zmask_acc = InMemoryDataAccessor(obj_table.zmask)
ti.click('generate_zmasks')
files = export_3d_patches_with_focus_metrics(
Path(where_output) / '3d_patches',
stack.get_one_channel_data(patches_channel),
- zmask_meta,
+ obj_table.zmask_meta,
prefix=fstem,
rescale_clip=0.0,
make_3d=True,
@@ -75,7 +82,7 @@ def export_patch_focus_metrics_from_multichannel_zstack(
files = export_patches_from_zstack(
Path(where_output) / '2d_patches',
stack.get_one_channel_data(patches_channel),
- zmask_meta,
+ obj_table.zmask_meta,
prefix=fstem,
draw_bounding_box=True,
rescale_clip=0.0,
@@ -88,11 +95,11 @@ def export_patch_focus_metrics_from_multichannel_zstack(
return {
'pixel_model_id': px_model.model_id,
'input_filepath': input_zstack_path,
- 'number_of_objects': len(zmask_meta),
+ 'number_of_objects': len(obj_table.zmask_meta),
'success': True,
'timer_results': ti.events,
'dataframe': df,
- 'interm': interm,
+ 'interm': obj_table.interm,
}
if __name__ == '__main__':
diff --git a/extensions/chaeo/params.py b/extensions/chaeo/params.py
new file mode 100644
index 00000000..02a500ef
--- /dev/null
+++ b/extensions/chaeo/params.py
@@ -0,0 +1,24 @@
+from typing import Dict, List, Union
+
+from pydantic import BaseModel
+
+class PatchParams(BaseModel):
+ draw_bounding_box: bool = False
+ draw_contour: bool = False
+ draw_mask: bool = False
+ rescale_clip: float = 0.001
+ focus_metric: str = 'max_sobel'
+ rgb_overlay_channels: List[int] = (None, None, None),
+ rgb_overlay_weights: List[float] = (1.0, 1.0, 1.0)
+
+class AnnotatedZStackParams(BaseModel):
+ draw_label: bool = False
+
+class ZMaskExportParams(BaseModel):
+ pixel_probabilities: bool = False
+ patches_3d: Union[PatchParams, None] = None
+ patches_2d_for_annotation: Union[PatchParams, None] = None
+ patches_2d_for_training: Union[PatchParams, None] = None
+ patch_masks: bool = False
+ annotated_z_stack: Union[AnnotatedZStackParams, None] = None
+
diff --git a/extensions/chaeo/workflows.py b/extensions/chaeo/workflows.py
index 3231ef6b..49383756 100644
--- a/extensions/chaeo/workflows.py
+++ b/extensions/chaeo/workflows.py
@@ -4,6 +4,7 @@ from uuid import uuid4
import numpy as np
import pandas as pd
+
from skimage.measure import label, regionprops_table
from skimage.morphology import dilation
from sklearn.model_selection import train_test_split
@@ -11,9 +12,10 @@ from sklearn.model_selection import train_test_split
from extensions.chaeo.accessors import MonoPatchStack
from extensions.chaeo.annotators import draw_boxes_on_3d_image
from extensions.chaeo.models import PatchStackObjectClassifier
+from extensions.chaeo.params import ZMaskExportParams
from extensions.chaeo.process import mask_largest_object
from extensions.chaeo.products import export_patches_from_zstack, export_patch_masks_from_zstack, export_multichannel_patches_from_zstack, get_patches_from_zmask_meta, get_patch_masks_from_zmask_meta
-from extensions.chaeo.zmask import build_zmask_from_object_mask, project_stack_from_focal_points
+from extensions.chaeo.zmask import project_stack_from_focal_points, ZMaskObjectTable
from extensions.ilastik.models import IlastikPixelClassifierModel
from model_server.accessors import generate_file_accessor, InMemoryDataAccessor, write_accessor_data_to_file
@@ -21,18 +23,232 @@ from model_server.models import Model
from model_server.process import rescale
from model_server.workflows import Timer
-def get_zmask_meta(
- input_file_path: str,
- ilastik_pixel_classifier: IlastikPixelClassifierModel,
- segmentation_channel: int,
- pxmap_threshold: float,
- pxmap_foreground_channel: int = 0,
- zmask_zindex: int = None,
- zmask_clip: int = None,
- zmask_filters: Dict = None,
- zmask_type: str = 'boxes',
- **kwargs,
-) -> tuple:
+# def get_zmask_meta(
+# input_file_path: str,
+# ilastik_pixel_classifier: IlastikPixelClassifierModel,
+# segmentation_channel: int,
+# pxmap_threshold: float,
+# pxmap_foreground_channel: int = 0,
+# zmask_zindex: int = None,
+# zmask_clip: int = None,
+# zmask_filters: Dict = None,
+# zmask_type: str = 'boxes',
+# **kwargs,
+# ) -> tuple:
+# ti = Timer()
+# stack = generate_file_accessor(Path(input_file_path))
+# fstem = Path(input_file_path).stem
+# ti.click('file_input')
+#
+# # MIP if no zmask z-index is given, then classify pixels
+# if isinstance(zmask_zindex, int):
+# assert 0 < zmask_zindex < stack.nz
+# zmask_data = stack.get_one_channel_data(channel=segmentation_channel).data[:, :, :, zmask_zindex]
+# else:
+# zmask_data = stack.get_one_channel_data(channel=segmentation_channel).data.max(axis=-1, keepdims=True)
+# if zmask_clip:
+# zmask_data = rescale(zmask_data, zmask_clip)
+# mip = InMemoryDataAccessor(
+# zmask_data,
+# )
+# pxmap, _ = ilastik_pixel_classifier.infer(mip)
+# ti.click('infer_pixel_probability')
+#
+# obmask = InMemoryDataAccessor(
+# pxmap.data > pxmap_threshold
+# )
+# ti.click('threshold_pixel_mask')
+#
+# # make zmask
+# obj_table = ZMaskObjectTable(
+# obmask.get_one_channel_data(pxmap_foreground_channel),
+# stack.get_one_channel_data(segmentation_channel),
+# mask_type=zmask_type,
+# filters=zmask_filters,
+# expand_box_by=kwargs['zmask_expand_box_by'],
+# )
+# ti.click('generate_zmasks')
+#
+# # record pixel scale
+# obj_table.df['pixel_scale_in_micrometers'] = float(stack.pixel_scale_in_micrometers.get('X'))
+#
+# return ti, stack, fstem, obmask, pxmap, obj_table
+
+
+# # called by batch runners
+# def export_patches_from_multichannel_zstack(
+# input_file_path: str,
+# output_folder_path: str,
+# models: List[Model],
+# pxmap_threshold: float,
+# pxmap_foreground_channel: int,
+# segmentation_channel: int,
+# patches_channel: int,
+# zmask_zindex: int = None, # None for MIP,
+# zmask_clip: int = None,
+# zmask_type: str = 'boxes',
+# zmask_filters: Dict = None,
+# zmask_expand_box_by: int = None,
+# export_pixel_probabilities=True,
+# export_2d_patches_for_training=True,
+# export_2d_patches_for_annotation=True,
+# draw_bounding_box_on_2d_patch=True,
+# draw_contour_on_2d_patch=False,
+# draw_mask_on_2d_patch=False,
+# export_3d_patches=True,
+# export_annotated_zstack=True,
+# draw_label_on_zstack=False,
+# export_patch_masks=True,
+# rgb_overlay_channels=(None, None, None),
+# rgb_overlay_weights=(1.0, 1.0, 1.0),
+# ) -> Dict:
+# pixel_classifier = models[0]
+#
+# # ti, stack, fstem, obmask, pxmap, obj_table = get_zmask_meta(
+# # input_file_path,
+# # pixel_classifier,
+# # segmentation_channel,
+# # pxmap_threshold,
+# # pxmap_foreground_channel=pxmap_foreground_channel,
+# # zmask_zindex=zmask_zindex,
+# # zmask_clip=zmask_clip,
+# # zmask_expand_box_by=zmask_expand_box_by,
+# # zmask_filters=zmask_filters,
+# # zmask_type=zmask_type,
+# # )
+#
+# # obj_table = ZMaskObjectTable(
+# # obmask.get_one_channel_data(pxmap_foreground_channel),
+# # stack.get_one_channel_data(segmentation_channel),
+# # mask_type=zmask_type,
+# # filters=zmask_filters,
+# # expand_box_by=kwargs['zmask_expand_box_by'],
+# # )
+#
+# if export_pixel_probabilities:
+# write_accessor_data_to_file(
+# Path(output_folder_path) / 'pixel_probabilities' / (fstem + '.tif'),
+# pxmap
+# )
+# ti.click('export_pixel_probability')
+#
+# if export_3d_patches and len(zmask_meta) > 0:
+# files = export_patches_from_zstack(
+# Path(output_folder_path) / '3d_patches',
+# stack.get_one_channel_data(patches_channel),
+# zmask_meta,
+# prefix=fstem,
+# draw_bounding_box=False,
+# rescale_clip=0.001,
+# make_3d=True,
+# )
+# ti.click('export_3d_patches')
+#
+# if export_2d_patches_for_annotation and len(zmask_meta) > 0:
+# files = export_multichannel_patches_from_zstack(
+# Path(output_folder_path) / '2d_patches_annotation',
+# stack,
+# zmask_meta,
+# prefix=fstem,
+# rescale_clip=0.001,
+# make_3d=False,
+# focus_metric='max_sobel',
+# ch_white=patches_channel,
+# ch_rgb_overlay=rgb_overlay_channels,
+# draw_bounding_box=draw_bounding_box_on_2d_patch,
+# bounding_box_channel=1,
+# bounding_box_linewidth=2,
+# draw_contour=draw_contour_on_2d_patch,
+# draw_mask=draw_mask_on_2d_patch,
+# overlay_gain=rgb_overlay_weights,
+# )
+# df_patches = pd.DataFrame(files)
+# ti.click('export_2d_patches')
+# # associate 2d patches, dropping labeled objects that were not exported as patches
+# df = pd.merge(df, df_patches, left_index=True, right_on='df_index').drop(columns='df_index')
+# # prepopulate patch UUID
+# df['patch_id'] = df.apply(lambda _: uuid4(), axis=1)
+#
+# if export_2d_patches_for_training and len(zmask_meta) > 0:
+# files = export_multichannel_patches_from_zstack(
+# Path(output_folder_path) / '2d_patches_training',
+# stack.get_one_channel_data(patches_channel),
+# zmask_meta,
+# prefix=fstem,
+# rescale_clip=0.001,
+# make_3d=False,
+# focus_metric='max_sobel',
+# )
+# ti.click('export_2d_patches')
+#
+# if export_patch_masks and len(zmask_meta) > 0:
+# files = export_patch_masks_from_zstack(
+# Path(output_folder_path) / 'patch_masks',
+# stack.get_one_channel_data(patches_channel),
+# zmask_meta,
+# prefix=fstem,
+# )
+#
+# if export_annotated_zstack:
+# annotated = InMemoryDataAccessor(
+# draw_boxes_on_3d_image(
+# stack.get_one_channel_data(patches_channel).data,
+# zmask_meta,
+# add_label=draw_label_on_zstack,
+# )
+# )
+# write_accessor_data_to_file(
+# Path(output_folder_path) / 'annotated_zstacks' / (fstem + '.tif'),
+# annotated
+# )
+# ti.click('export_annotated_zstack')
+#
+# # generate multichannel projection from label centroids
+# dff = df[df['keeper']]
+# if len(zmask_meta) > 0:
+# interm['projected'] = project_stack_from_focal_points(
+# dff['centroid-0'].to_numpy(),
+# dff['centroid-1'].to_numpy(),
+# dff['zi'].to_numpy(),
+# stack,
+# degree=4,
+# )
+# else: # else just return MIP
+# interm['projected'] = stack.data.max(axis=-1)
+#
+# return {
+# 'pixel_model_id': pixel_classifier.model_id,
+# 'input_filepath': input_file_path,
+# 'number_of_objects': len(zmask_meta),
+# 'pixeL_scale_in_micrometers': stack.pixel_scale_in_micrometers,
+# 'success': True,
+# 'timer_results': ti.events,
+# 'dataframe': df[df['keeper'] == True],
+# 'interm': interm,
+# }
+
+def infer_object_map_from_zstack(
+ input_file_path: str,
+ output_folder_path: str,
+ models: List[Model],
+ pxmap_foreground_channel: int,
+ pxmap_threshold: float,
+ segmentation_channel: int,
+ patches_channel: int,
+ zmask_zindex: int = None, # None for MIP,
+ zmask_clip: int = None,
+ zmask_type: str = 'boxes',
+ zmask_filters: Dict = None,
+ # zmask_expand_box_by: int = None,
+ exports: ZMaskExportParams = None,
+ **kwargs,
+) -> Dict:
+ assert len(models) == 2
+ pixel_classifier = models[0]
+ assert isinstance(pixel_classifier, IlastikPixelClassifierModel)
+ object_classifier = models[1]
+ assert isinstance(object_classifier, PatchStackObjectClassifier)
+
ti = Timer()
stack = generate_file_accessor(Path(input_file_path))
fstem = Path(input_file_path).stem
@@ -49,16 +265,23 @@ def get_zmask_meta(
mip = InMemoryDataAccessor(
zmask_data,
)
- pxmap, _ = ilastik_pixel_classifier.infer(mip)
+ pxmap, _ = pixel_classifier.infer(mip)
ti.click('infer_pixel_probability')
+ if exports.pixel_probabilities:
+ write_accessor_data_to_file(
+ Path(output_folder_path) / 'pixel_probabilities' / (fstem + '.tif'),
+ pxmap
+ )
+ ti.click('export_pixel_probability')
+
obmask = InMemoryDataAccessor(
pxmap.data > pxmap_threshold
)
ti.click('threshold_pixel_mask')
# make zmask
- zmask, zmask_meta, df, interm = build_zmask_from_object_mask(
+ obj_table = ZMaskObjectTable(
obmask.get_one_channel_data(pxmap_foreground_channel),
stack.get_one_channel_data(segmentation_channel),
mask_type=zmask_type,
@@ -68,194 +291,26 @@ def get_zmask_meta(
ti.click('generate_zmasks')
# record pixel scale
- df['pixel_scale_in_micrometers'] = float(stack.pixel_scale_in_micrometers.get('X'))
-
- return ti, stack, fstem, obmask, pxmap, zmask, zmask_meta, df, interm
-
-
-# TODO: unpack and validate inputs
-def export_patches_from_multichannel_zstack(
- input_file_path: str,
- output_folder_path: str,
- models: List[Model],
- pxmap_threshold: float,
- pxmap_foreground_channel: int,
- segmentation_channel: int,
- patches_channel: int,
- zmask_zindex: int = None, # None for MIP,
- zmask_clip: int = None,
- zmask_type: str = 'boxes',
- zmask_filters: Dict = None,
- zmask_expand_box_by: int = None,
- export_pixel_probabilities=True,
- export_2d_patches_for_training=True,
- export_2d_patches_for_annotation=True,
- draw_bounding_box_on_2d_patch=True,
- draw_contour_on_2d_patch=False,
- draw_mask_on_2d_patch=False,
- export_3d_patches=True,
- export_annotated_zstack=True,
- draw_label_on_zstack=False,
- export_patch_masks=True,
- rgb_overlay_channels=(None, None, None),
- rgb_overlay_weights=(1.0, 1.0, 1.0),
-) -> Dict:
- pixel_classifier = models[0]
-
- ti, stack, fstem, obmask, pxmap, zmask, zmask_meta, df, interm = get_zmask_meta(
- input_file_path,
- pixel_classifier,
- segmentation_channel,
- pxmap_threshold,
- pxmap_foreground_channel=pxmap_foreground_channel,
- zmask_zindex=zmask_zindex,
- zmask_clip=zmask_clip,
- zmask_expand_box_by=zmask_expand_box_by,
- zmask_filters=zmask_filters,
- zmask_type=zmask_type,
- )
-
- if export_pixel_probabilities:
- write_accessor_data_to_file(
- Path(output_folder_path) / 'pixel_probabilities' / (fstem + '.tif'),
- pxmap
- )
- ti.click('export_pixel_probability')
-
- if export_3d_patches and len(zmask_meta) > 0:
- files = export_patches_from_zstack(
- Path(output_folder_path) / '3d_patches',
- stack.get_one_channel_data(patches_channel),
- zmask_meta,
- prefix=fstem,
- draw_bounding_box=False,
- rescale_clip=0.001,
- make_3d=True,
- )
- ti.click('export_3d_patches')
-
- if export_2d_patches_for_annotation and len(zmask_meta) > 0:
- files = export_multichannel_patches_from_zstack(
- Path(output_folder_path) / '2d_patches_annotation',
- stack,
- zmask_meta,
- prefix=fstem,
- rescale_clip=0.001,
- make_3d=False,
- focus_metric='max_sobel',
- ch_white=patches_channel,
- ch_rgb_overlay=rgb_overlay_channels,
- draw_bounding_box=draw_bounding_box_on_2d_patch,
- bounding_box_channel=1,
- bounding_box_linewidth=2,
- draw_contour=draw_contour_on_2d_patch,
- draw_mask=draw_mask_on_2d_patch,
- overlay_gain=rgb_overlay_weights,
- )
- df_patches = pd.DataFrame(files)
- ti.click('export_2d_patches')
- # associate 2d patches, dropping labeled objects that were not exported as patches
- df = pd.merge(df, df_patches, left_index=True, right_on='df_index').drop(columns='df_index')
- # prepopulate patch UUID
- df['patch_id'] = df.apply(lambda _: uuid4(), axis=1)
-
- if export_2d_patches_for_training and len(zmask_meta) > 0:
- files = export_multichannel_patches_from_zstack(
- Path(output_folder_path) / '2d_patches_training',
- stack.get_one_channel_data(patches_channel),
- zmask_meta,
- prefix=fstem,
- rescale_clip=0.001,
- make_3d=False,
- focus_metric='max_sobel',
- )
- ti.click('export_2d_patches')
-
- if export_patch_masks and len(zmask_meta) > 0:
- files = export_patch_masks_from_zstack(
- Path(output_folder_path) / 'patch_masks',
- stack.get_one_channel_data(patches_channel),
- zmask_meta,
- prefix=fstem,
- )
-
- if export_annotated_zstack:
- annotated = InMemoryDataAccessor(
- draw_boxes_on_3d_image(
- stack.get_one_channel_data(patches_channel).data,
- zmask_meta,
- add_label=draw_label_on_zstack,
- )
- )
- write_accessor_data_to_file(
- Path(output_folder_path) / 'annotated_zstacks' / (fstem + '.tif'),
- annotated
- )
- ti.click('export_annotated_zstack')
-
- # generate multichannel projection from label centroids
- dff = df[df['keeper']]
- if len(zmask_meta) > 0:
- interm['projected'] = project_stack_from_focal_points(
- dff['centroid-0'].to_numpy(),
- dff['centroid-1'].to_numpy(),
- dff['zi'].to_numpy(),
- stack,
- degree=4,
- )
- else: # else just return MIP
- interm['projected'] = stack.data.max(axis=-1)
-
- return {
- 'pixel_model_id': pixel_classifier.model_id,
- 'input_filepath': input_file_path,
- 'number_of_objects': len(zmask_meta),
- 'pixeL_scale_in_micrometers': stack.pixel_scale_in_micrometers,
- 'success': True,
- 'timer_results': ti.events,
- 'dataframe': df[df['keeper'] == True],
- 'interm': interm,
- }
-
-def infer_object_map_from_zstack(
- input_file_path: str,
- output_folder_path: str,
- models: List[Model],
- pxmap_threshold: float,
- pxmap_foreground_channel: int,
- segmentation_channel: int,
- patches_channel: int,
- zmask_zindex: int = None, # None for MIP,
- zmask_clip: int = None,
- zmask_type: str = 'boxes',
- zmask_filters: Dict = None,
- # zmask_expand_box_by: int = None,
- **kwargs,
-) -> Dict:
- assert len(models) == 2
- pixel_classifier = models[0]
- assert isinstance(pixel_classifier, IlastikPixelClassifierModel)
- object_classifier = models[1]
- assert isinstance(object_classifier, PatchStackObjectClassifier)
-
- ti, stack, fstem, obmask, pxmap, zmask, zmask_meta, df, interm = get_zmask_meta(
- input_file_path,
- pixel_classifier,
- segmentation_channel,
- pxmap_threshold,
- pxmap_foreground_channel=pxmap_foreground_channel,
- zmask_zindex=zmask_zindex,
- zmask_clip=zmask_clip,
- # zmask_expand_box_by=zmask_expand_box_by,
- zmask_filters=zmask_filters,
- zmask_type=zmask_type,
- **kwargs
- )
+ obj_table.df['pixel_scale_in_micrometers'] = float(stack.pixel_scale_in_micrometers.get('X'))
+
+ # ti, stack, fstem, obmask, pxmap, obj_table = get_zmask_meta(
+ # input_file_path,
+ # pixel_classifier,
+ # segmentation_channel,
+ # pxmap_threshold,
+ # pxmap_foreground_channel=pxmap_foreground_channel,
+ # zmask_zindex=zmask_zindex,
+ # zmask_clip=zmask_clip,
+ # # zmask_expand_box_by=zmask_expand_box_by,
+ # zmask_filters=zmask_filters,
+ # zmask_type=zmask_type,
+ # **kwargs
+ # )
# extract patches to accessor
patches_acc = get_patches_from_zmask_meta(
stack.get_one_channel_data(patches_channel),
- zmask_meta,
+ obj_table.zmask_meta,
rescale_clip=zmask_clip,
make_3d=False,
focus_metric='max_sobel',
@@ -265,22 +320,22 @@ def infer_object_map_from_zstack(
# extract masks
patch_masks_acc = get_patch_masks_from_zmask_meta(
stack,
- zmask_meta,
+ obj_table.zmask_meta,
**kwargs
)
# send patches and mask stacks to object classifier
result_acc, _ = object_classifier.infer(patches_acc, patch_masks_acc)
- labels_map = interm['label_map']
+ labels_map = obj_table.interm['label_map']
output_map = np.zeros(labels_map.shape, dtype=labels_map.dtype)
- assert labels_map.shape == interm['label_map'].shape
- assert labels_map.dtype == interm['label_map'].dtype
+ assert labels_map.shape == obj_table.get_label_map().shape
+ assert labels_map.dtype == obj_table.get_label_map().dtype
# assign labels to object map:
meta = []
- for ii in range(0, len(zmask_meta)):
- object_id = zmask_meta[ii]['info'].label
+ for ii in range(0, len(obj_table.zmask_meta)):
+ object_id = obj_table.zmask_meta[ii]['info'].label
result_patch = mask_largest_object(result_acc.iat(ii))
object_class = np.unique(result_patch)[1]
output_map[labels_map == object_id] = object_class
@@ -293,6 +348,51 @@ def infer_object_map_from_zstack(
)
ti.click('export_object_classes')
+ if exports.patches_3d:
+ obj_table.export_3d_patches(
+ Path(output_folder_path) / '3d_patches',
+ fstem,
+ patches_channel,
+ exports.patches_3d
+ )
+ ti.click('export_3d_patches')
+
+ if exports.patches_2d_for_annotation:
+ obj_table.export_2d_patches_for_annotation(
+ Path(output_folder_path) / '2d_patches_annotation',
+ fstem,
+ patches_channel,
+ exports.patches_2d_for_annotation
+ )
+ ti.click('export_2d_patches_for_annotation')
+
+ if exports.patches_2d_for_training:
+ obj_table.export_2d_patches_for_training(
+ Path(output_folder_path) / '2d_patches_training',
+ fstem,
+ patches_channel,
+ exports.patches_2d_for_training
+ )
+ ti.click('export_2d_patches_for_training')
+
+ if exports.patch_masks:
+ obj_table.export_patch_masks(
+ Path(output_folder_path) / 'patch_masks',
+ fstem,
+ patches_channel,
+ exports.patch_masks
+ )
+
+ if exports.annotated_z_stack:
+ obj_table.export_annotated_zstack(
+ Path(output_folder_path) / 'patch_masks',
+ fstem,
+ patches_channel,
+ exports.annotated_z_stack
+ )
+ ti.click('export_annotated_zstack')
+
+
return {
'timer_results': ti.events,
'dataframe': pd.DataFrame(meta),
diff --git a/extensions/chaeo/zmask.py b/extensions/chaeo/zmask.py
index e9967e70..24ff80ef 100644
--- a/extensions/chaeo/zmask.py
+++ b/extensions/chaeo/zmask.py
@@ -1,3 +1,5 @@
+from uuid import uuid4
+
import numpy as np
import pandas as pd
@@ -5,7 +7,149 @@ from skimage.measure import find_contours, label, regionprops_table
from sklearn.preprocessing import PolynomialFeatures
from sklearn.linear_model import LinearRegression
-from model_server.accessors import GenericImageDataAccessor
+from extensions.chaeo.annotators import draw_boxes_on_3d_image
+from extensions.chaeo.products import export_patches_from_zstack, export_multichannel_patches_from_zstack, export_patch_masks_from_zstack
+from extensions.chaeo.params import ZMaskExportParams
+from model_server.accessors import GenericImageDataAccessor, InMemoryDataAccessor, write_accessor_data_to_file
+
+
+
+class ZMaskObjectTable(object):
+
+ def __init__(
+ self,
+ acc_mask: GenericImageDataAccessor,
+ acc_raw: GenericImageDataAccessor,
+ filters=None,
+ mask_type='contours',
+ expand_box_by=(0, 0),
+ ):
+ self.zmask, self.zmask_meta, self.df, self.interm = build_zmask_from_object_mask(
+ acc_mask,
+ acc_raw,
+ filters=filters,
+ mask_type=mask_type,
+ expand_box_by=expand_box_by
+ )
+ self.acc_raw = acc_raw
+ self.count = len(self.zmask_meta)
+
+ def get_label_map(self):
+ return self.interm.lamap
+
+ def get_argmax(self):
+ return self.interm.argmax
+
+ def export_3d_patches(self, where, prefix, channel, params: ZMaskExportParams):
+ if not self.count:
+ return
+ files = export_patches_from_zstack(
+ where,
+ self.acc_raw.get_one_channel_data(channel),
+ self.zmask_meta,
+ prefix=prefix,
+ draw_bounding_box=params.draw_bounding_box,
+ rescale_clip=params.rescale_clip,
+ make_3d=True,
+ )
+
+ def export_2d_patches_for_annotation(self, where, prefix, channel, params: ZMaskExportParams):
+ if not self.count:
+ return
+ files = export_multichannel_patches_from_zstack(
+ where,
+ self.acc_raw.get_one_channel_data(channel),
+ self.zmask_meta,
+ prefix=prefix,
+ draw_bounding_box=params.draw_bounding_box,
+ rescale_clip=params.rescale_clip,
+ make_3d=False,
+ focus_metric=params.focus_metric,
+ ch_white=channel,
+ ch_rgb_overlay=params.rgb_overlay_channels,
+ bounding_box_channel=1,
+ bounding_box_linewidth=2,
+ draw_contour=params.draw_contour,
+ draw_mask=params.draw_mask,
+ overlay_gain=params.rgb_overlay_weights,
+ )
+ df_patches = pd.DataFrame(files)
+ self.df = pd.merge(self.df, df_patches, left_index=True, right_on='df_index').drop(columns='df_index')
+ self.df['patch_id'] = self.df.apply(lambda _: uuid4(), axis=1)
+
+ def export_2d_patches_for_training(self, where, prefix, channel, params: ZMaskExportParams):
+ if not self.count:
+ return
+ files = export_multichannel_patches_from_zstack(
+ where,
+ self.acc_raw.get_one_channel_data(channel),
+ self.zmask_meta,
+ prefix=prefix,
+ draw_bounding_box=params.draw_bounding_box,
+ rescale_clip=params.rescale_clip,
+ make_3d=False,
+ focus_metric=params.focus_metric,
+ ch_white=channel,
+ ch_rgb_overlay=params.rgb_overlay_channels,
+ bounding_box_channel=1,
+ bounding_box_linewidth=2,
+ draw_contour=params.draw_contour,
+ draw_mask=params.draw_mask,
+ overlay_gain=params.rgb_overlay_weights,
+ )
+ df_patches = pd.DataFrame(files)
+ self.df = pd.merge(self.df, df_patches, left_index=True, right_on='df_index').drop(columns='df_index')
+ self.df['patch_id'] = self.df.apply(lambda _: uuid4(), axis=1)
+
+ def export_2d_patches_for_annotation(self, where, prefix, channel, params: ZMaskExportParams):
+ if not self.count:
+ return
+ files = export_multichannel_patches_from_zstack(
+ where,
+ self.acc_raw.get_one_channel_data(channel),
+ self.zmask_meta,
+ prefix=prefix,
+ rescale_clip=params.rescale_clip,
+ make_3d=False,
+ focus_metric=params.focus_metric,
+ )
+
+ def export_patch_masks(self, where, prefix, channel, params: ZMaskExportParams):
+ if not self.count:
+ return
+ files = export_patch_masks_from_zstack(
+ where,
+ self.acc_raw.get_one_channel_data(channel),
+ self.zmask_meta,
+ prefix=prefix,
+ )
+
+ def export_annotated_zstack(self, where, prefix, channel, params: ZMaskExportParams):
+ annotated = InMemoryDataAccessor(
+ draw_boxes_on_3d_image(
+ self.acc_raw.get_one_channel_data(channel).data,
+ self.zmask_meta,
+ add_label=params.draw_label,
+ )
+ )
+ write_accessor_data_to_file(
+ where / 'annotated_zstacks' / (prefix + '.tif'),
+ annotated
+ )
+
+ def get_multichannel_projection(self):
+ dff = self.df[self.df['keeper']]
+ if self.count:
+ projected = project_stack_from_focal_points(
+ dff['centroid-0'].to_numpy(),
+ dff['centroid-1'].to_numpy(),
+ dff['zi'].to_numpy(),
+ self.acc_raw,
+ degree=4,
+ )
+ else: # else just return MIP
+ projected = self.acc_raw.data.max(axis=-1)
+ return projected
def build_zmask_from_object_mask(
obmask: GenericImageDataAccessor,
diff --git a/extensions/ilastik/models.py b/extensions/ilastik/models.py
index 1413456f..21278461 100644
--- a/extensions/ilastik/models.py
+++ b/extensions/ilastik/models.py
@@ -53,6 +53,7 @@ class IlastikImageToImageModel(ImageToImageModel):
class IlastikPixelClassifierModel(IlastikImageToImageModel):
model_id = 'ilastik_pixel_classification'
+ operations = ['segment', ]
@staticmethod
def get_workflow():
@@ -77,35 +78,40 @@ class IlastikPixelClassifierModel(IlastikImageToImageModel):
)
return InMemoryDataAccessor(data=yxcz), {'success': True}
-class IlastikObjectClassifierFromPixelPredictionsModel(IlastikImageToImageModel):
- model_id = 'ilastik_object_classification_from_pixel_predictions'
-
- @staticmethod
- def get_workflow():
- from ilastik.workflows.objectClassification.objectClassificationWorkflow import ObjectClassificationWorkflowPrediction
- return ObjectClassificationWorkflowPrediction
-
- def infer(self, input_img: GenericImageDataAccessor, pxmap_img: GenericImageDataAccessor) -> (np.ndarray, dict):
- tagged_input_data = vigra.taggedView(input_img.data, 'yxcz')
- tagged_pxmap_data = vigra.taggedView(pxmap_img.data, 'yxcz')
-
- dsi = [
- {
- 'Raw Data': self.PreloadedArrayDatasetInfo(preloaded_array=tagged_input_data),
- 'Prediction Maps': self.PreloadedArrayDatasetInfo(preloaded_array=tagged_pxmap_data),
- }
- ]
-
- obmaps = self.shell.workflow.batchProcessingApplet.run_export(dsi, export_to_array=True) # [z x h x w x n]
-
- assert (len(obmaps) == 1, 'ilastik generated more than one object map')
-
- yxcz = np.moveaxis(
- obmaps[0],
- [1, 2, 3, 0],
- [0, 1, 2, 3]
+ def segment(self, input_img, thresh, channel):
+ return InMemoryDataAccessor(
+ self.infer(input_img).data[:, :, channel, :] > thresh
)
- return InMemoryDataAccessor(data=yxcz), {'success': True}
+
+# class IlastikObjectClassifierFromPixelPredictionsModel(IlastikImageToImageModel):
+# model_id = 'ilastik_object_classification_from_pixel_predictions'
+#
+# @staticmethod
+# def get_workflow():
+# from ilastik.workflows.objectClassification.objectClassificationWorkflow import ObjectClassificationWorkflowPrediction
+# return ObjectClassificationWorkflowPrediction
+#
+# def infer(self, input_img: GenericImageDataAccessor, pxmap_img: GenericImageDataAccessor) -> (np.ndarray, dict):
+# tagged_input_data = vigra.taggedView(input_img.data, 'yxcz')
+# tagged_pxmap_data = vigra.taggedView(pxmap_img.data, 'yxcz')
+#
+# dsi = [
+# {
+# 'Raw Data': self.PreloadedArrayDatasetInfo(preloaded_array=tagged_input_data),
+# 'Prediction Maps': self.PreloadedArrayDatasetInfo(preloaded_array=tagged_pxmap_data),
+# }
+# ]
+#
+# obmaps = self.shell.workflow.batchProcessingApplet.run_export(dsi, export_to_array=True) # [z x h x w x n]
+#
+# assert (len(obmaps) == 1, 'ilastik generated more than one object map')
+#
+# yxcz = np.moveaxis(
+# obmaps[0],
+# [1, 2, 3, 0],
+# [0, 1, 2, 3]
+# )
+# return InMemoryDataAccessor(data=yxcz), {'success': True}
class IlastikObjectClassifierFromSegmentationModel(IlastikImageToImageModel):
diff --git a/extensions/ilastik/router.py b/extensions/ilastik/router.py
index 14ec3258..4d9a8b28 100644
--- a/extensions/ilastik/router.py
+++ b/extensions/ilastik/router.py
@@ -35,7 +35,11 @@ def load_ilastik_model(model_class: ilm.IlastikImageToImageModel, project_file:
)
return {'model_id': result}
-@router.put('/px/load/')
+# @router.put('/px/load/')
+# def load_px_model(project_file: str, duplicate: bool = True) -> dict:
+# return load_ilastik_model(ilm.IlastikPixelClassifierModel, project_file, duplicate=duplicate)
+
+@router.put('/seg/load/')
def load_px_model(project_file: str, duplicate: bool = True) -> dict:
return load_ilastik_model(ilm.IlastikPixelClassifierModel, project_file, duplicate=duplicate)
--
GitLab