def get_target(self, index: int) -> Dict[str, Any]:
"""
Returns the instance segmentation target
"""
target = self.parse_json(index)
annotations = []
for annotation in target["annotations"]:
if "polygon" not in annotation and "complex_polygon" not in annotation:
print(
f"Warning: missing polygon in annotation {self.annotations_path[index]}"
)
# Extract the sequences of coordinates from the polygon annotation
annotation_type: str = "polygon" if "polygon" in annotation else "complex_polygon"
sequences = convert_polygons_to_sequences(
annotation[annotation_type]["path"],
height=target["height"],
width=target["width"],
)
# Compute the bbox of the polygon
x_coords = [s[0::2] for s in sequences]
y_coords = [s[1::2] for s in sequences]
min_x = np.min([np.min(x_coord) for x_coord in x_coords])
min_y = np.min([np.min(y_coord) for y_coord in y_coords])
max_x = np.max([np.max(x_coord) for x_coord in x_coords])
max_y = np.max([np.max(y_coord) for y_coord in y_coords])
w = max_x - min_x + 1
h = max_y - min_y + 1
# Compute the area of the polygon
# TODO fix with addictive/subtractive paths in complex polygons
poly_area = np.sum([
polygon_area(x_coord, y_coord)
for x_coord, y_coord in zip(x_coords, y_coords)
])
# Create and append the new entry for this annotation
annotations.append({
"category_id":
self.classes.index(annotation["name"]),
"segmentation":
sequences,
"bbox": [min_x, min_y, w, h],
"area":
poly_area,
})
target["annotations"] = annotations
return target
def _map_annotation(self, index: int):
"""See superclass for documentation
Notes
-----
The return value is a dict with the following fields:
image_id : int
Index of the image inside the dataset
original_filename: str
The path to the image on the file system
annotations : list[Dict]
List of annotations, where each annotation is a dict with:
category_id : int
The single label of the image selected.
segmentation : ndarray(1,)
Array of points [x,y,x,y,x,y ...] composing the polygon enclosing the object
bbox : ndarray(1,)
Coordinates of the bounding box enclosing the instance as [x, y, w, h]
area : float
Area of the polygon
"""
with self.annotations_path[index].open() as f:
annotations = json.load(f)["annotations"]
# Filter out unused classes
if self.classes is not None:
annotations = [a for a in annotations if a["name"] in self.classes]
target = []
for annotation in annotations:
assert "name" in annotation
assert "polygon" in annotation
# Extract the sequences of coordinates from the polygon annotation
sequences = convert_polygons_to_sequences(annotation["polygon"]["path"])
# Discard polygons with less than three points
sequences[:] = [s for s in sequences if len(s) >= 6]
if not sequences:
continue
# Compute the bbox of the polygon
x_coords = [s[0::2] for s in sequences]
y_coords = [s[1::2] for s in sequences]
min_x = np.min([np.min(x_coord) for x_coord in x_coords])
min_y = np.min([np.min(y_coord) for y_coord in y_coords])
max_x = np.max([np.max(x_coord) for x_coord in x_coords])
max_y = np.max([np.max(y_coord) for y_coord in y_coords])
w = max_x - min_x + 1
h = max_y - min_y + 1
bbox_area = w * h
# Compute the area of the polygon
poly_area = np.sum(
[polygon_area(x_coord, y_coord) for x_coord, y_coord in zip(x_coords, y_coords)]
)
assert poly_area <= bbox_area
# Create and append the new entry for this annotation
target.append(
{
"category_id": self.classes.index(annotation["name"]),
"segmentation": sequences,
"bbox": [min_x, min_y, w, h],
"area": poly_area,
}
)
return {
"image_id": index,
"original_filename": self.images_path[index],
"annotations": target,
}
def __getitem__(self, index: int):
"""
Notes
-----
The return value is a dict with the following fields:
image_id : int
Index of the image inside the dataset
image_path: str
The path to the image on the file system
labels : tensor(n)
The class label of each one of the instances
masks : tensor(n, H, W)
Segmentation mask of each one of the instances
boxes : tensor(n, 4)
Coordinates of the bounding box enclosing the instances as [x, y, x, y]
area : float
Area in pixels of each one of the instances
"""
img = load_pil_image(self.images_path[index])
target = self.parse_json(index)
annotations = []
for annotation in target["annotations"]:
if "polygon" not in annotation and "complex_polygon" not in annotation:
print(
f"Warning: missing polygon in annotation {self.annotations_path[index]}"
)
# Extract the sequences of coordinates from the polygon annotation
annotation_type = "polygon" if "polygon" in annotation else "complex_polygon"
sequences = convert_polygons_to_sequences(
annotation[annotation_type]["path"],
height=target["height"],
width=target["width"],
)
# Compute the bbox of the polygon
x_coords = [s[0::2] for s in sequences]
y_coords = [s[1::2] for s in sequences]
min_x = np.min([np.min(x_coord) for x_coord in x_coords])
min_y = np.min([np.min(y_coord) for y_coord in y_coords])
max_x = np.max([np.max(x_coord) for x_coord in x_coords])
max_y = np.max([np.max(y_coord) for y_coord in y_coords])
w = max_x - min_x + 1
h = max_y - min_y + 1
# Compute the area of the polygon
# TODO fix with addictive/subtractive paths in complex polygons
poly_area = np.sum([
polygon_area(x_coord, y_coord)
for x_coord, y_coord in zip(x_coords, y_coords)
])
# Create and append the new entry for this annotation
annotations.append({
"category_id":
self.classes.index(annotation["name"]),
"segmentation":
sequences,
"bbox": [min_x, min_y, w, h],
"area":
poly_area,
})
target["annotations"] = annotations
img, target = self.convert_polygons(img, target)
if self.transform is not None:
img, target = self.transform(img, target)
return img, target