fp2 / src /fiberpose /domain.py
vikba's picture
Upload folder using huggingface_hub
405af39 verified
Raw History Blame Contribute Delete
26.2 kB
"""Core data models for configuration, batches, frame outputs, and fiber geometry."""
from __future__ import annotations
import json
from dataclasses import dataclass, field, asdict
from copy import copy
from typing import Any, Dict, Iterable, List, Optional
import os
import imageio.v3 as iio
import numpy as np
import logging
from tifffile import TiffFile
logger = logging.getLogger(__name__)
@dataclass
class AppConfig:
# --- Model & Hardware ---
"""Store configuration values for app."""
model_path: str = ""
use_gpu: bool = True
scale_label: str = "1x"
# --- Centerline Params ---
max_angle_deg: float = 30.0
ridge_sigma: float = 1.0
ridge_low: float = 0.01
ridge_high: float = 0.4
min_fragment_centerline_length_px: float = 10.0
measurement_centerline_length_trim_fraction: float = 0.05
grad_quality_trim_fraction: float = 0.0
strict_centerlines: bool = False
keep_segments_without_measurement_centerlines: bool = False
# --- Segmentation Params ---
segmentation_mode: str = "watershed"
merge_expansion_factor: float = 0.7
segmentation_expansion_factor: float = 1.0
watershed_grad_threshold: float = 0.3
# Retained for DT-based void/pore analysis; watershed uses the gradient threshold above.
seg_threshold: int = 1
min_segment_foreground_mean: float = 0.5
# --- Analysis & Physical Units ---
sample_spacing: float = 5.0
# --- DT gradient quality view ---
# Signed error breakpoints for s = |grad(dt)| - 1:
# [p0, p1, p2, p3] with p0 < p1 <= 0 <= p2 < p3.
grad_quality_points: list[float] = field(
default_factory=lambda: [-0.35, -0.08, 0.08, 0.35]
)
# --- Save/export defaults (single image, folder batch, and stack exports) ---
save_options: dict = field(
default_factory=lambda: {
"dt": True,
"foreground": True,
"labels": True,
"overlay": True,
"diameters": True,
"orientations": True,
"outline": False,
"tables": True,
"image_format": "png",
"collision": "suffix",
"write_manifest": True,
}
)
def save_to_json(self, filepath: str):
"""Persist to json using the configured export conventions.
Args:
filepath (str): Filesystem location used by this operation."""
with open(filepath, 'w') as f:
json.dump(asdict(self), f, indent=4)
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "AppConfig":
"""Normalize a config mapping into AppConfig."""
if not isinstance(data, dict):
raise ValueError("Config JSON root must be an object.")
data = dict(data)
valid_keys = set(cls.__dataclass_fields__)
unknown_keys = sorted(set(data.keys()) - valid_keys)
if unknown_keys:
raise ValueError(f"Unknown config key(s): {', '.join(unknown_keys)}")
return cls(**data)
@classmethod
def load_from_json(cls, filepath: str) -> "AppConfig":
try:
with open(filepath, "r", encoding="utf-8") as f:
data = json.load(f)
return cls.from_dict(data)
except FileNotFoundError:
logger.info("Config file not found: %s. Using defaults.", filepath)
return cls()
@dataclass
class SampleProperties:
"""Per-loaded-sample physical calibration used by analysis and UI."""
pixel_size_x: float = 1.0
pixel_size_y: float = 1.0
voxel_size_z: float = 1.0
unit: str = "px"
source: str = "default"
source_path: str = ""
@staticmethod
def _as_positive(value: Any, default: float = 1.0) -> float:
if isinstance(value, (tuple, list)) and len(value) == 2:
try:
value = float(value[0]) / float(value[1])
except (TypeError, ValueError, ZeroDivisionError):
return float(default)
try:
out = float(value)
except (TypeError, ValueError):
return float(default)
if out <= 0 or not np.isfinite(out):
return float(default)
return out
@classmethod
def defaults(cls) -> "SampleProperties":
return cls(
pixel_size_x=1.0,
pixel_size_y=1.0,
voxel_size_z=1.0,
unit="px",
source="default",
source_path="",
)
@classmethod
def from_image_metadata(
cls,
metadata: Optional[Dict[str, Any]],
*,
source_path: str = "",
) -> "SampleProperties":
meta = metadata or {}
px_raw = meta.get("pixel_width")
py_raw = meta.get("pixel_height")
pz_raw = meta.get("voxel_depth")
unit_raw = meta.get("unit")
has_px = px_raw is not None and cls._as_positive(px_raw, default=-1.0) > 0
has_py = py_raw is not None and cls._as_positive(py_raw, default=-1.0) > 0
has_pz = pz_raw is not None and cls._as_positive(pz_raw, default=-1.0) > 0
px = cls._as_positive(px_raw) if has_px else None
py = cls._as_positive(py_raw) if has_py else None
pz = cls._as_positive(pz_raw) if has_pz else 1.0
if px is None and py is None:
px = 1.0
py = 1.0
elif px is None:
px = py
elif py is None:
py = px
unit_txt = str(unit_raw).strip() if unit_raw is not None else ""
unit = unit_txt if unit_txt else "px"
has_metadata = bool(has_px or has_py or has_pz or unit_txt)
return cls(
pixel_size_x=float(px),
pixel_size_y=float(py),
voxel_size_z=float(pz),
unit=unit,
source="metadata" if has_metadata else "default",
source_path=str(source_path or meta.get("source_path") or ""),
)
def effective_pixel_size_xy(self) -> float:
x = self._as_positive(self.pixel_size_x)
y = self._as_positive(self.pixel_size_y)
return 0.5 * (x + y)
def spacing_zyx(self) -> tuple[float, float, float]:
return (
self._as_positive(self.voxel_size_z),
self._as_positive(self.pixel_size_y),
self._as_positive(self.pixel_size_x),
)
@dataclass
class ImageBatch:
"""Container for an ordered batch of frames plus lightweight metadata."""
frames: List[np.ndarray]
metadata: Dict[str, Any] = field(default_factory=dict)
def __len__(self) -> int:
"""Report how many items are currently tracked by this container.
Returns:
int: Number of items currently available in the container."""
return len(self.frames)
@property
def kind(self) -> str:
"""Returns the kind of loaded info: 'Image', 'Folder', or 'Stack'."""
return self.metadata.get("source_type", "Image")
@property
def name(self) -> str:
"""Returns the base name of the file or folder."""
path = self.metadata.get("source_path", "")
if not path:
return "Unknown"
return os.path.basename(os.path.normpath(path))
@staticmethod
def _extract_voxel_sizes(path: str) -> Dict[str, Any]:
"""Read physical calibration from TIFF tags and common metadata keys."""
sizes = {"pixel_width": None, "pixel_height": None, "voxel_depth": None, "unit": None}
def _positive(value: Any) -> Optional[float]:
if value is None:
return None
if isinstance(value, (tuple, list)) and len(value) == 2:
try:
value = float(value[0]) / float(value[1])
except (TypeError, ValueError, ZeroDivisionError):
return None
try:
out = float(value)
except (TypeError, ValueError):
return None
if out <= 0 or not np.isfinite(out):
return None
return out
def _unit_scale_to_um(unit_value: Any) -> Optional[float]:
if unit_value is None:
return None
token = str(getattr(unit_value, "name", unit_value)).strip().lower()
if token in {"inch", "in", "inches"}:
return 25400.0
if token in {"centimeter", "centimetre", "cm"}:
return 10000.0
if token in {"millimeter", "millimetre", "mm"}:
return 1000.0
if token in {"resunit.inch", "resunit.centimeter", "resunit.centimetre"}:
if "inch" in token:
return 25400.0
return 10000.0
try:
token_num = int(unit_value)
except (TypeError, ValueError):
return None
if token_num == 2:
return 25400.0
if token_num == 3:
return 10000.0
return None
try:
meta = iio.immeta(path)
except Exception:
meta = {}
sizes["voxel_depth"] = (
_positive(meta.get("spacing"))
or _positive(meta.get("PhysicalSizeZ"))
)
sizes["unit"] = str(
meta.get("unit")
or meta.get("PhysicalSizeXUnit")
or meta.get("PhysicalSizeYUnit")
or meta.get("PhysicalSizeZUnit")
or ""
).strip() or None
sizes["pixel_width"] = (
_positive(meta.get("pixel_width"))
or _positive(meta.get("PhysicalSizeX"))
)
sizes["pixel_height"] = (
_positive(meta.get("pixel_height"))
or _positive(meta.get("PhysicalSizeY"))
)
if (
(sizes["pixel_width"] is None or sizes["pixel_height"] is None)
and str(path).lower().endswith((".tif", ".tiff"))
):
with TiffFile(path) as tif:
tags = tif.pages[0].tags
x_res = _positive(tags["XResolution"].value if "XResolution" in tags else None)
y_res = _positive(tags["YResolution"].value if "YResolution" in tags else None)
res_unit = tags["ResolutionUnit"].value if "ResolutionUnit" in tags else None
unit_scale = _unit_scale_to_um(res_unit)
if sizes["pixel_width"] is None and x_res is not None:
sizes["pixel_width"] = (1.0 / x_res) * (unit_scale or 1.0)
if sizes["pixel_height"] is None and y_res is not None:
sizes["pixel_height"] = (1.0 / y_res) * (unit_scale or 1.0)
if sizes["unit"] is None and unit_scale is not None:
sizes["unit"] = "um"
# Minimal fallback for non-TIFF metadata formats.
if sizes["pixel_width"] is None or sizes["pixel_height"] is None:
res = meta.get("resolution")
rx = None
ry = None
if isinstance(res, (tuple, list)) and len(res) >= 2:
rx = _positive(res[0])
ry = _positive(res[1])
elif isinstance(res, dict):
rx = _positive(res.get("x") or res.get("X"))
ry = _positive(res.get("y") or res.get("Y"))
unit_scale = _unit_scale_to_um(
meta.get("resolution_unit")
or meta.get("ResolutionUnit")
or meta.get("resolutionunit")
)
if sizes["pixel_width"] is None and rx is not None:
sizes["pixel_width"] = (1.0 / rx) * (unit_scale or 1.0)
if sizes["pixel_height"] is None and ry is not None:
sizes["pixel_height"] = (1.0 / ry) * (unit_scale or 1.0)
if sizes["unit"] is None and unit_scale is not None:
sizes["unit"] = "um"
return sizes
@staticmethod
def _normalize_stack(data: np.ndarray) -> List[np.ndarray]:
if data.ndim == 2:
return [data]
if data.ndim == 3:
if data.shape[2] in (3, 4):
return [data]
return [data[z] for z in range(data.shape[0])]
if data.ndim == 4:
return [data[z] for z in range(data.shape[0])]
raise ValueError("Unsupported image dimensions for batch loading.")
@classmethod
def from_path(cls, path: str) -> "ImageBatch":
data = iio.imread(path)
frames = cls._normalize_stack(data)
source_type = "Image" if len(frames) == 1 else "Stack"
filenames = [os.path.basename(path)] * len(frames)
# --- Extract physical sizes ---
voxel_sizes = cls._extract_voxel_sizes(path)
metadata = {
"source_type": source_type,
"source_path": path,
"filenames": filenames,
"z_indices": list(range(len(frames))),
**voxel_sizes # Merge the extracted sizes into the metadata dictionary
}
return cls(frames=frames, metadata=metadata)
@classmethod
def from_folder(
cls, folder: str, extensions: Optional[Iterable[str]] = None
) -> "ImageBatch":
entries = sorted(os.listdir(folder))
ext_set = {ext.lower() for ext in extensions} if extensions else None
frames: List[np.ndarray] = []
filenames: List[str] = []
first_valid_path = None
for name in entries:
path = os.path.join(folder, name)
if not os.path.isfile(path):
continue
if ext_set is not None:
if os.path.splitext(name)[1].lower() not in ext_set:
continue
try:
data = iio.imread(path)
if first_valid_path is None:
first_valid_path = path # Save the first image to extract metadata
except Exception:
continue
image_frames = cls._normalize_stack(data)
for image_frame in image_frames:
frames.append(image_frame)
filenames.append(name)
# --- Extract physical sizes from the first image in the folder ---
voxel_sizes = {}
if first_valid_path:
voxel_sizes = cls._extract_voxel_sizes(first_valid_path)
metadata = {
"source_type": "Folder",
"source_path": folder,
"filenames": filenames,
"z_indices": list(range(len(frames))),
**voxel_sizes
}
return cls(frames=frames, metadata=metadata)
@classmethod
def from_source(
cls, path: str, extensions: Optional[Iterable[str]] = None
) -> "ImageBatch":
if os.path.isdir(path):
return cls.from_folder(path, extensions=extensions)
if os.path.isfile(path):
return cls.from_path(path)
raise FileNotFoundError(f"Source path does not exist: {path}")
@dataclass
class BatchResult:
"""Capture outputs and metadata produced by batch."""
frame_results: List[Frame]
def frame_at(self, index: int) -> Frame:
return self.frame_results[index]
def save_all(self, output_dir: str, export_config: Dict[str, Any] | None = None) -> Dict[str, Any]:
"""Persist all using the configured export conventions.
Args:
output_dir (str): Filesystem location used by this operation.
export_config (Dict[str, Any] | None): Configuration values controlling runtime behavior.
Returns:
Dict[str, Any]: Mapping of computed keys and values."""
from .io_export import BatchResultExporter
exporter = BatchResultExporter(export_config=export_config)
return exporter.save_all(self, output_dir, export_config=export_config)
@dataclass
class Frame:
"""Flat, stage-aware container for a single frame's processing state."""
# Identity
z_index: int = 0
frame_id: str = ""
# Stage flags
has_prediction: bool = False
has_segmentation: bool = False
has_measurements: bool = False
prediction_sig: Optional[str] = None
segmentation_sig: Optional[str] = None
measurement_sig: Optional[str] = None
error_message: Optional[str] = None
# Source and prediction buffers
orig_img: Optional[np.ndarray] = None
orig_img_scaled: Optional[np.ndarray] = None
scale_factor: float = 1.0
dt: Optional[np.ndarray] = None
dt_u8: Optional[np.ndarray] = None
dt_normalized: Optional[np.ndarray] = None
foreground_prob: Optional[np.ndarray] = None
lapl_img: Optional[np.ndarray] = None
gx: Optional[np.ndarray] = None
gy: Optional[np.ndarray] = None
# Segmentation products
labels_2d: Optional[np.ndarray] = None
# Fragments that own a label; never modified after segmentation, so snapshots share them.
fragments: Optional[List[FiberFragment]] = None
# Measurement outputs
samples_table: Optional[Any] = None
fiber_summary: Optional[Dict[str, Any]] = None
measurement_samples: Optional[Any] = None
drawing_samples: Optional[Any] = None
thickness_units: Optional[np.ndarray] = None
porosity: Optional[float] = None
pore_diams_units: Optional[np.ndarray] = None
void_mask: Optional[np.ndarray] = None
pore_dt: Optional[np.ndarray] = None
pores: Optional[Any] = None
# Images built by `render.render`, keyed by view; emptied whenever any stage is cleared.
views: Dict[str, Optional[np.ndarray]] = field(default_factory=dict, repr=False)
@classmethod
def empty(
cls,
*,
frame_id: str = "",
z_index: int = 0,
orig_img: Optional[np.ndarray] = None,
) -> "Frame":
return cls(frame_id=str(frame_id), z_index=int(z_index), orig_img=orig_img)
def clear_prediction(self) -> None:
self.has_prediction = False
self.dt = None
self.dt_u8 = None
self.dt_normalized = None
self.foreground_prob = None
self.lapl_img = None
self.gx = None
self.gy = None
self.orig_img_scaled = None
self.scale_factor = 1.0
self.prediction_sig = None
self.error_message = None
self.clear_segmentation()
def clear_segmentation(self) -> None:
self.has_segmentation = False
self.labels_2d = None
self.fragments = None
self.segmentation_sig = None
self.clear_measurements()
def clear_measurements(self) -> None:
self.has_measurements = False
self.samples_table = None
self.fiber_summary = None
self.measurement_samples = None
self.drawing_samples = None
self.thickness_units = None
self.porosity = None
self.pore_diams_units = None
self.void_mask = None
self.pore_dt = None
self.pores = None
self.measurement_sig = None
self.views = {}
def set_prediction_outputs(
self,
*,
dt: Optional[np.ndarray],
gx: Optional[np.ndarray],
gy: Optional[np.ndarray],
dt_u8: Optional[np.ndarray],
dt_normalized: Optional[np.ndarray],
foreground_prob: Optional[np.ndarray],
lapl_img: Optional[np.ndarray],
orig_img_scaled: Optional[np.ndarray],
scale_factor: float,
) -> None:
self.dt = dt
self.gx = gx
self.gy = gy
self.dt_u8 = dt_u8
self.dt_normalized = dt_normalized
self.foreground_prob = foreground_prob
self.lapl_img = lapl_img
self.orig_img_scaled = orig_img_scaled
self.scale_factor = float(scale_factor)
self.has_prediction = self.dt is not None
self.error_message = None
self.clear_segmentation()
def set_segmentation_outputs(
self,
*,
labels_2d: np.ndarray,
fragments: List[FiberFragment],
) -> None:
self.labels_2d = labels_2d
self.fragments = fragments
self.has_segmentation = True
self.error_message = None
self.clear_measurements()
def set_measurement_outputs(
self,
*,
samples_table: Optional[Any],
fiber_summary: Optional[Dict[str, Any]],
measurement_samples: Optional[Any],
drawing_samples: Optional[Any],
thickness_units: Optional[np.ndarray],
porosity: Optional[float],
pore_diams_units: Optional[np.ndarray],
void_mask: Optional[np.ndarray],
pore_dt: Optional[np.ndarray],
pores: Optional[Any],
) -> None:
"""Update measurement outputs and synchronize dependent runtime state."""
self.samples_table = samples_table
self.fiber_summary = fiber_summary
self.measurement_samples = measurement_samples
self.drawing_samples = drawing_samples
self.thickness_units = thickness_units
self.porosity = porosity
self.pore_diams_units = pore_diams_units
self.void_mask = void_mask
self.pore_dt = pore_dt
self.pores = pores
self.has_measurements = True
self.error_message = None
def fail(self, message: str) -> None:
self.error_message = str(message)
def snapshot(
self,
*,
frame_id: Optional[str] = None,
z_index: Optional[int] = None,
) -> "Frame":
clone: Frame = copy(self)
def _clone_array(value: Any) -> Any:
if isinstance(value, np.ndarray):
return value.copy()
return value
clone.dt = _clone_array(clone.dt)
clone.dt_u8 = _clone_array(clone.dt_u8)
clone.dt_normalized = _clone_array(clone.dt_normalized)
clone.foreground_prob = _clone_array(clone.foreground_prob)
clone.lapl_img = _clone_array(clone.lapl_img)
clone.gx = _clone_array(clone.gx)
clone.gy = _clone_array(clone.gy)
clone.labels_2d = _clone_array(clone.labels_2d)
clone.samples_table = _clone_array(clone.samples_table)
clone.thickness_units = _clone_array(clone.thickness_units)
clone.pore_diams_units = _clone_array(clone.pore_diams_units)
clone.void_mask = _clone_array(clone.void_mask)
clone.pore_dt = _clone_array(clone.pore_dt)
# Cached images are never modified in place, so the clone may share them.
clone.views = dict(clone.views)
if clone.fiber_summary is not None:
clone.fiber_summary = dict(clone.fiber_summary)
if frame_id is not None:
clone.frame_id = str(frame_id)
if z_index is not None:
clone.z_index = int(z_index)
return clone
@dataclass
class FiberFragment:
"""Unified 2D fragment carrying both segmentation and measurement centerlines."""
fragment_id: int
segment_centerlines_xy: list[np.ndarray]
segment_radii: Optional[list[np.ndarray]] = None
parent_fiber_id: int = -1
measurement_centerlines_xy: list[np.ndarray] = field(default_factory=list)
total_measurement_centerline_length_px: float = 0.0
grad_quality_score: float = float("nan")
@staticmethod
def _normalize_centerline(points_raw: np.ndarray, label: str) -> np.ndarray:
points_xy = np.asarray(points_raw, dtype=np.float32)
if points_xy.ndim != 2 or points_xy.shape[1] != 2:
raise ValueError(f"{label} must be an array of shape (N, 2).")
if points_xy.shape[0] < 2:
raise ValueError(f"{label} must contain at least two points.")
return points_xy
@staticmethod
def _normalize_radii(radii_raw: Optional[np.ndarray], n_points: int, label: str) -> np.ndarray:
if radii_raw is None:
return np.zeros(n_points, dtype=np.float32)
radii = np.asarray(radii_raw, dtype=np.float32).reshape(-1)
if radii.shape[0] != n_points:
raise ValueError(f"{label} length must match centerline length.")
return radii
@staticmethod
def _centerline_length(points_xy: np.ndarray) -> float:
if points_xy.shape[0] < 2:
return 0.0
return float(np.linalg.norm(np.diff(points_xy, axis=0), axis=1).sum())
def recompute_total_measurement_centerline_length_px(self) -> float:
self.total_measurement_centerline_length_px = float(
sum(self._centerline_length(points_xy) for points_xy in self.measurement_centerlines_xy)
)
return self.total_measurement_centerline_length_px
def __post_init__(self) -> None:
if not self.segment_centerlines_xy:
raise ValueError("FiberFragment must contain at least one segmentation centerline.")
centerlines_xy = [
self._normalize_centerline(points_raw, "Segmentation centerline")
for points_raw in self.segment_centerlines_xy
]
if not self.segment_radii:
centerlines_radii = [
np.zeros(points_xy.shape[0], dtype=np.float32) for points_xy in centerlines_xy
]
else:
if len(self.segment_radii) != len(centerlines_xy):
raise ValueError("segment_radii length must match segment_centerlines_xy length.")
centerlines_radii = [
self._normalize_radii(radii_raw, points_xy.shape[0], "Segmentation radii")
for points_xy, radii_raw in zip(centerlines_xy, self.segment_radii)
]
pairs = list(zip(centerlines_xy, centerlines_radii))
pairs.sort(key=lambda pair: self._centerline_length(pair[0]), reverse=True)
self.segment_centerlines_xy = [points_xy for points_xy, _ in pairs]
self.segment_radii = [radii for _, radii in pairs]
self.measurement_centerlines_xy = [
self._normalize_centerline(points_raw, "Measurement centerline")
for points_raw in (self.measurement_centerlines_xy or [])
]
self.recompute_total_measurement_centerline_length_px()
self.fragment_id = int(self.fragment_id)
self.parent_fiber_id = int(self.parent_fiber_id)
@property
def has_measurement_centerline(self) -> bool:
return bool(self.measurement_centerlines_xy)