core

core utils

source

load_tif

 load_tif (fpath:Union[str,pathlib.Path], chnls_last=True)

Loads a geotiff file, transpose its channels if necessary, and return the image data as a numpy array.

Parameters

  • fpath (str|pathlib.Path): The file path of the image to be loaded.
  • chnls_last (bool, optional): If set to True, the function will return the image data in the format (height, width, channels) (h, w, c). If set to False, the function will return the image data in the format (channels, height, width) (c, h, w). Default value is True.

Returns

  • numpy.ndarray: The image data as a NumPy array. The shape of the array depends on the value of chnls_last.
    • If chnls_last is True, the returned array will have the shape (height, width, channels).
    • If chnls_last is False, the returned array will have the shape (channels, height, width).
Type Default Details
fpath Union file path of the image to be loaded
chnls_last bool True transpose from c,h,w to h,w,c (assume c,h,w on rio.read input)
Returns ndarray ndarray will return either h,w,c or c,h,w depending on chnls_last is True or False

Explore loading geotiffs

from torchgeo.datasets import EuroSAT100
import fastai.vision.all as fv
# load a sample geotiff dataset
sat_path = fv.untar_data(EuroSAT100.url); sat_path
Path('/home/butch2/.fastai/data/EuroSAT100')
sat_path.ls()
(#1) [Path('/home/butch2/.fastai/data/EuroSAT100/images')]
sat_images = fv.get_image_files(sat_path); len(sat_images)
100
sat_image = sat_images[10]; sat_image
Path('/home/butch2/.fastai/data/EuroSAT100/images/remote_sensing/otherDatasets/sentinel_2/tif/River/River_2160.tif')
data = load_tif(sat_image) # channels last is the default
(type(data), data.shape) # height,width, channels
(numpy.ndarray, (64, 64, 13))
data = load_tif(sat_image, chnls_last=False)
data.shape # channels, height, width
(13, 64, 64)

source

open_tif

 open_tif (fn:Union[str,pathlib.Path], chnls:[<class'int'>]=None,
           chnls_last:bool=True, cls=<class 'torch.Tensor'>)

Creates a tensor from a TIFF image file.

Parameters

  • fn (str): The path to the TIFF image file.
  • chnls (list of int, optional): The channels to load from the image. Default to None, which loads all channels
  • chnls_last (bool, optional): If True, the channel dimension is the last dimension. Default to True.
  • cls (class, optional): The class of the tensor to create. Defaults to torch.Tensor

Returns

  • Tensor: A pytorch tensor created from the TIFF image file.
Type Default Details
fn Union The path to the TIFF image file
chnls [<class ‘int’>] None The channels to load from the image
chnls_last bool True If True, the channel dimension is the last dimension
cls _TensorMeta Tensor The class of the tensor to create
Returns Tensor A tensor array of the image

Loading geotiffs as Tensors

t1 = open_tif(sat_image)
(type(t1), t1.shape)
(torch.Tensor, torch.Size([64, 64, 13]))

source

GeoTensorImage

 GeoTensorImage (x, chnls_first=False)

A TensorImage for geotiffs

Can contain more than 3 channels (e.g. 12 band sentinel-2 images) as well as a bigger range of values (16-bit vs 8-bit) than your typical image formats

Attributes

  • chnls_first (bool): If True, the format of the tensor is c,h,w else its h,w,c where c = channels, h = height, w = width

Methods

  • create(data, chnls, chnls_first): (classmethod) create a GeoTensorImage from a file or numpy array
  • show(chnls,scaler,bright, ctx): display a GeoTensorImage
GeoTensorImage.create?
Signature:
GeoTensorImage.create(
    data: Union[pathlib.Path, str, numpy.ndarray],
    chnls: [<class 'int'>] = None,
    chnls_first: bool = True,
) -> fastai.torch_core.TensorImage
Docstring:
(class method) Returns an instance of the class with the input image data formatted appropriately. 
If input is Path/str and ends with "tif", opens it using open_tif function. 
If input is np.ndarray, transposes it if chnls_first. 
Otherwise, input is used as is.
## Parameters:
- `data` (Union[Path, str, np.ndarray]): The image data that the class method is called on.
- `chnls` ([int]): An optional list of channels to filter. Defaults to None, which does no filtering.
- `chnls_first` (bool, optional): A boolean flag indicating whether the channels are in the first or last 
                                  dimension of the input array. Defaults to True.
## Returns
- GeoTensorImage: An instance of the GeoTensorImage class with appropriate image formatting.
File:      /tmp/ipykernel_21397/1243907552.py
Type:      method

Creating GeoTensorImages from a file

geo_img = GeoTensorImage.create(sat_image)
geo_img
GeoTensorImage: torch.Size([13, 64, 64])
geo_img.show?
Signature:
geo_img.show(
    chnls: [<class 'int'>] = [3, 2, 1],
    scaler: float = 1.0,
    bright: float = 1.0,
    ctx: matplotlib.axes._axes.Axes = None,
) -> matplotlib.image.AxesImage
Docstring:
Shows the GeoTensorImage object in a matplotlib figure.
Args:
    chnls: A list of channel indices to show. The default is [3, 2, 1].
    scaler: A float value to scale the image by. The default is 1.0.
    bright: A float value to brighten the image by. The default is 1.0.
    ctx: A matplotlib context object. The default is None.
Returns:
    The matplotlib Axes image.
File:      /tmp/ipykernel_21397/1243907552.py
Type:      method

Showing GeoTensorImages

geo_img.show();

geo_img.show(bright=1.7);

geo_img.show(scaler=0.7);

geo_img.show(scaler=0.7, bright=1.7);

fig,ax = plt.subplots(1, 2, figsize=(10,10))
geo_img.show(ctx=ax[0])
geo_img.show(ctx=ax[1], scaler=0.4)
plt.show()