Skip to content

Instantly share code, notes, and snippets.

@braised-babbage
Created May 17, 2020 01:08
Show Gist options
  • Save braised-babbage/1ab51f0e306e06d461f80e221a2e76f5 to your computer and use it in GitHub Desktop.
Save braised-babbage/1ab51f0e306e06d461f80e221a2e76f5 to your computer and use it in GitHub Desktop.
from collections import defaultdict
import numpy as np
from typing import Dict, Any, List, Union, Callable, Set, Iterable, Iterator, Optional, Tuple
from dataclasses import dataclass
@dataclass
class Image:
""" Representation of an image, along with metadata. """
data: np.ndarray
header: Dict[str, Any]
@property
def shape(self) -> Tuple[int]:
return self.data.shape
def metadata(self, *keys: str) -> Tuple[Any]:
return tuple(self.header.get(key, None) for key in keys)
class ImageSet:
""" A queryable set of images. """
def __init__(self, *imagesets: Union[Image, Iterable[Image]]):
self.images = []
for ims in imagesets:
if isinstance(ims, Image):
self.images.append(ims)
else:
for im in ims:
self.images.append(im)
def __iter__(self) -> Iterator[Image]:
return iter(self.images)
def __len__(self) -> int:
return len(self.images)
def query(self, **kwargs) -> 'ImageSet':
""" Get a subset of images with metadata satisfying certain conditions. """
def matches(image: Image) -> bool:
return image.metadata(*kwargs.keys()) == tuple(kwargs.values())
return ImageSet([img for img in self if matches(img)])
def __repr__(self):
return f"<ImageSet of size {len(self)}>"
@dataclass
class Transform:
""" A transformation mapping an ImageSet to an ImageSet. """
transform_op: Callable[..., Union[Image, Iterable[Image]]]
partition_keys: Optional[List[str]] = None
def __call__(self, images: ImageSet) -> ImageSet:
def partition_value(image: Image):
if self.partition_keys is not None:
return image.metadata(*self.partition_keys)
else:
return id(image)
partition = defaultdict(list)
for image in images:
partition[partition_value(image)].append(image)
result = ImageSet([self.transform_op(*images) for images in partition.values()])
return result
# Examples
a = Image(data=np.random.rand(10), header={'type': 'raw', 'segment': 10})
b = Image(data=np.random.rand(10), header={'type': 'hr', 'segment': 10})
c = Image(data=np.random.rand(10), header={'type': 'raw', 'segment': 12})
d = Image(data=np.random.rand(10), header={'type': 'hr', 'segment': 12})
images = ImageSet(a,b,c,d)
raw = images.query(type='raw')
doubled = ImageSet(images, images)
def increment_data(img: Image) -> Image:
return Image(data=img.data + 1.0, header=img.header)
def subtract_raw(*imgs: Image) -> List[Image]:
raws = [img for img in imgs if img.metadata('type') == ('raw',)]
assert len(raws) == 1
raw, = raws
return [Image(data=img.data-raw.data, header=img.header)
for img in imgs if img.metadata('type') == ('hr',)]
IncrementTransform = Transform(transform_op=increment_data)
SubtractRaw = Transform(transform_op=subtract_raw, partition_keys=['segment'])
reduced = SubtractRaw(images)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment