Skip to content

Commit

Permalink
Utils klasöründeki kodlar kısaltıldı.
Browse files Browse the repository at this point in the history
  • Loading branch information
@kadirnar
kadirnar committed Mar 8, 2022
1 parent d174f89 commit 30122a1
Show file tree
Hide file tree
Showing 12 changed files with 15 additions and 2,688 deletions.
Binary file removed demo_data/prediction_visual.png
View file
Binary file not shown.
199 changes: 1 addition & 198 deletions sahi_lite/annotation.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,11 +6,6 @@

import numpy as np
from sahi_lite.utils.coco import CocoAnnotation, CocoPrediction
from sahi_lite.utils.cv import (
get_bbox_from_bool_mask,
get_bool_mask_from_coco_segmentation,
get_coco_segmentation_from_bool_mask,
)
from sahi_lite.utils.shapely import ShapelyAnnotation


Expand Down Expand Up @@ -107,175 +102,6 @@ def __repr__(self):
return f"Category: <id: {self.id}, name: {self.name}>"


class Mask:
@classmethod
def from_float_mask(
cls,
mask,
full_shape=None,
mask_threshold: float = 0.5,
shift_amount: list = [0, 0],
):
"""
Args:
mask: np.ndarray of np.float elements
Mask values between 0 and 1 (should have a shape of height*width)
mask_threshold: float
Value to threshold mask pixels between 0 and 1
shift_amount: List
To shift the box and mask predictions from sliced image
to full sized image, should be in the form of [shift_x, shift_y]
full_shape: List
Size of the full image after shifting, should be in the form of [height, width]
"""
bool_mask = mask > mask_threshold
return cls(
bool_mask=bool_mask,
shift_amount=shift_amount,
full_shape=full_shape,
)

@classmethod
def from_coco_segmentation(
cls,
segmentation,
full_shape=None,
shift_amount: list = [0, 0],
):
"""
Init Mask from coco segmentation representation.
Args:
segmentation : List[List]
[
[x1, y1, x2, y2, x3, y3, ...],
[x1, y1, x2, y2, x3, y3, ...],
...
]
full_shape: List
Size of the full image, should be in the form of [height, width]
shift_amount: List
To shift the box and mask predictions from sliced image to full
sized image, should be in the form of [shift_x, shift_y]
"""
# confirm full_shape is given
assert full_shape is not None, "full_shape must be provided"

bool_mask = get_bool_mask_from_coco_segmentation(segmentation, height=full_shape[0], width=full_shape[1])
return cls(
bool_mask=bool_mask,
shift_amount=shift_amount,
full_shape=full_shape,
)

def __init__(
self,
bool_mask=None,
full_shape=None,
shift_amount: list = [0, 0],
):
"""
Args:
bool_mask: np.ndarray with bool elements
2D mask of object, should have a shape of height*width
full_shape: List
Size of the full image, should be in the form of [height, width]
shift_amount: List
To shift the box and mask predictions from sliced image to full
sized image, should be in the form of [shift_x, shift_y]
"""

if len(bool_mask) > 0:
has_bool_mask = True
else:
has_bool_mask = False

if has_bool_mask:
self.bool_mask = bool_mask.astype(bool)
else:
self.bool_mask = None

self.shift_x = shift_amount[0]
self.shift_y = shift_amount[1]

if full_shape:
self.full_shape_height = full_shape[0]
self.full_shape_width = full_shape[1]
elif has_bool_mask:
self.full_shape_height = self.bool_mask.shape[0]
self.full_shape_width = self.bool_mask.shape[1]
else:
self.full_shape_height = None
self.full_shape_width = None

@property
def shape(self):
"""
Returns mask shape as [height, width]
"""
return [self.bool_mask.shape[0], self.bool_mask.shape[1]]

@property
def full_shape(self):
"""
Returns full mask shape after shifting as [height, width]
"""
return [self.full_shape_height, self.full_shape_width]

@property
def shift_amount(self):
"""
Returns the shift amount of the mask slice as [shift_x, shift_y]
"""
return [self.shift_x, self.shift_y]

def get_shifted_mask(self):
# Confirm full_shape is specified
assert (self.full_shape_height is not None) and (self.full_shape_width is not None), "full_shape is None"
# init full mask
mask_fullsized = np.full(
(
self.full_shape_height,
self.full_shape_width,
),
0,
dtype="float32",
)

# arrange starting ending indexes
starting_pixel = [self.shift_x, self.shift_y]
ending_pixel = [
min(starting_pixel[0] + self.bool_mask.shape[1], self.full_shape_width),
min(starting_pixel[1] + self.bool_mask.shape[0], self.full_shape_height),
]

# convert sliced mask to full mask
mask_fullsized[starting_pixel[1]: ending_pixel[1], starting_pixel[0]: ending_pixel[0]] = self.bool_mask[
: ending_pixel[1] -
starting_pixel[1],
: ending_pixel[0] -
starting_pixel[0]
]

return Mask(
mask_fullsized,
shift_amount=[0, 0],
full_shape=self.full_shape,
)

def to_coco_segmentation(self):
"""
Returns boolean mask as coco segmentation:
[
[x1, y1, x2, y2, x3, y3, ...],
[x1, y1, x2, y2, x3, y3, ...],
...
]
"""
coco_segmentation = get_coco_segmentation_from_bool_mask(self.bool_mask)
return coco_segmentation


class ObjectAnnotation:
"""
All about an annotation such as Mask, Category, BoundingBox.
Expand Down Expand Up @@ -348,10 +174,8 @@ def from_coco_segmentation(
To shift the box and mask predictions from sliced image to full
sized image, should be in the form of [shift_x, shift_y]
"""
bool_mask = get_bool_mask_from_coco_segmentation(segmentation, width=full_shape[1], height=full_shape[0])
return cls(
category_id=category_id,
bool_mask=bool_mask,
category_name=category_name,
shift_amount=shift_amount,
full_shape=full_shape,
Expand Down Expand Up @@ -459,12 +283,8 @@ def from_shapely_annotation(
To shift the box and mask predictions from sliced image to full
sized image, should be in the form of [shift_x, shift_y]
"""
bool_mask = get_bool_mask_from_coco_segmentation(
annotation.to_coco_segmentation(), width=full_shape[1], height=full_shape[0]
)
return cls(
category_id=category_id,
bool_mask=bool_mask,
category_name=category_name,
shift_amount=shift_amount,
full_shape=full_shape,
Expand Down Expand Up @@ -527,20 +347,7 @@ def __init__(
if (bbox is None) and (bool_mask is None):
raise ValueError("you must provide a bbox or bool_mask")

if bool_mask is not None:
self.mask = Mask(
bool_mask=bool_mask,
shift_amount=shift_amount,
full_shape=full_shape,
)
bbox_from_bool_mask = get_bbox_from_bool_mask(bool_mask)
# https://github.com/obss/sahi/issues/235
if bbox_from_bool_mask is not None:
bbox = bbox_from_bool_mask
else:
raise ValueError("Invalid boolean mask.")
else:
self.mask = None
self.mask = None

# make sure bbox coords lie inside [0, image_size]
xmin = max(bbox[0], 0)
Expand Down Expand Up @@ -645,10 +452,6 @@ def deepcopy(self):
"""
return copy.deepcopy(self)

@classmethod
def get_empty_mask(cls):
return Mask(bool_mask=None)

def get_shifted_object_annotation(self):
if self.mask:
return ObjectAnnotation(
Expand Down
Empty file added sahi_lite/postprocess/__init__.py
View file
Empty file.
1 change: 1 addition & 0 deletions sahi_lite/postprocess/legacy/__init__.py
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@

22 changes: 3 additions & 19 deletions sahi_lite/postprocess/utils.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@

import numpy as np
import torch
from sahi_lite.annotation import BoundingBox, Category, Mask
from sahi_lite.annotation import BoundingBox, Category
from sahi_lite.prediction import ObjectPrediction


Expand Down Expand Up @@ -156,17 +156,6 @@ def has_match(
return threshold_condition


def get_merged_mask(pred1: ObjectPrediction, pred2: ObjectPrediction) -> Mask:
mask1 = pred1.mask
mask2 = pred2.mask
union_mask = np.logical_or(mask1.bool_mask, mask2.bool_mask)
return Mask(
bool_mask=union_mask,
full_shape=mask1.full_shape,
shift_amount=mask1.shift_amount,
)


def get_merged_score(
pred1: ObjectPrediction,
pred2: ObjectPrediction,
Expand Down Expand Up @@ -197,13 +186,8 @@ def merge_object_prediction_pair(
merged_bbox: BoundingBox = get_merged_bbox(pred1, pred2)
merged_score: float = get_merged_score(pred1, pred2)
merged_category: Category = get_merged_category(pred1, pred2)
if pred1.mask and pred2.mask:
merged_mask: Mask = get_merged_mask(pred1, pred2)
bool_mask = merged_mask.bool_mask
full_shape = merged_mask.full_shape
else:
bool_mask = None
full_shape = None
bool_mask = None
full_shape = None
return ObjectPrediction(
bbox=merged_bbox.to_voc_bbox(),
score=merged_score,
Expand Down
38 changes: 1 addition & 37 deletions sahi_lite/prediction.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
from PIL import Image
from sahi_lite.annotation import ObjectAnnotation
from sahi_lite.utils.cv import read_image_as_pil, visualize_object_predictions
from sahi_lite.utils.file import Path
from pathlib import Path


class PredictionScore:
Expand Down Expand Up @@ -106,42 +106,6 @@ def get_shifted_object_prediction(self):
full_shape=None,
)

def to_coco_prediction(self, image_id=None):
"""
Returns sahi.utils.coco.CocoPrediction representation of ObjectAnnotation.
"""
if self.mask:
coco_prediction = CocoPrediction.from_coco_segmentation(
segmentation=self.mask.to_coco_segmentation(),
category_id=self.category.id,
category_name=self.category.name,
score=self.score.value,
image_id=image_id,
)
else:
coco_prediction = CocoPrediction.from_coco_bbox(
bbox=self.bbox.to_coco_bbox(),
category_id=self.category.id,
category_name=self.category.name,
score=self.score.value,
image_id=image_id,
)
return coco_prediction

def to_fiftyone_detection(self, image_height: int, image_width: int):
"""
Returns fiftyone.Detection representation of ObjectPrediction.
"""
try:
import fiftyone as fo
except ImportError:
raise ImportError('Please run "pip install -U fiftyone" to install fiftyone first for fiftyone conversion.')

x1, y1, x2, y2 = self.bbox.to_voc_bbox()
rel_box = [x1 / image_width, y1 / image_height, (x2 - x1) / image_width, (y2 - y1) / image_height]
fiftyone_detection = fo.Detection(label=self.category.name, bounding_box=rel_box, confidence=self.score.value)
return fiftyone_detection

def __repr__(self):
return f"""ObjectPrediction<
bbox: {self.bbox},
Expand Down
Loading

0 comments on commit 30122a1

Please sign in to comment.