Spaces:

hackvermin
/

Setellite_super_resolution

Paused

Setellite_super_resolution / worldstrat /src /datasources.py

Upload 26 files

8931de0 verified about 1 month ago

3.19 kB

import os
import torch
from glob import glob

# ------------------------------------------------------ #
#                  Sentinel2   - S2
#                  Level 2A   - L2A
#                  Level 1C   - L1C
# ------------------------------------------------------ #


# Source: https://custom-scripts.sentinel-hub.com/custom-scripts/sentinel-2/composites/
# NOTE: these are 1-indexed because GDAL is 1-indexed.
S2_ALL_12BANDS = {
    "true_color": [4, 3, 2],
    "false_color": [8, 4, 3],
    "swir": [12, 8, 4],
    "agriculture": [11, 8, 2],
    "geology": [12, 11, 2],
    "bathimetric": [4, 3, 1],
    "true_color_zero_index": [3, 2, 1],
}

S2_ALL_BANDS = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]

# ------------------------------------------------------ #
#                  SpaceNet7   - SN7
#                  PlanetScope - PS
# ------------------------------------------------------ #

PS_BANDS = [1, 2, 3]

METADATA_PATH = 'dataset/metadata.csv'
ROOT_SN7_DATA = os.path.join("data", "SN7")
ROOT_SN7_DATA_TRAIN = os.path.join(ROOT_SN7_DATA, "train")
SCENES_SN7 = glob(os.path.join(ROOT_SN7_DATA_TRAIN, "L15*"))

SN7_SUBDIRECTORIES = {"lr": "S2", "lrc": "S2_masks", "hr": "images"}
SN7_BANDS_TO_READ = {"lr": S2_ALL_12BANDS["true_color"], "lrc": None, "hr": PS_BANDS}


# Precomputed on training data by running: src/datasets.py
S2_SN7_MEAN = torch.Tensor(
    [
        0.0769,
        0.0906,
        0.1162,
        0.1318,
        0.1619,
        0.2275,
        0.2500,
        0.2588,
        0.2667,
        0.2684,
        0.2461,
        0.1874,
    ]
).to(torch.float64)

S2_SN7_STD = torch.Tensor(
    [
        0.1708,
        0.1672,
        0.1614,
        0.1677,
        0.1698,
        0.1513,
        0.1490,
        0.1515,
        0.1461,
        0.2365,
        0.1472,
        0.1418,
    ]
).to(torch.float64)

PS_MEAN = torch.Tensor([105.0, 92.0, 66.0]).to(torch.float64)  # RGB
PS_STD = torch.Tensor([58.2366, 47.1755, 44.9830]).to(torch.float64)
SN7_MAX_EXPECTED_HR_VALUE = 255


# ------------------------------------------------------ #
#   Satellite pour l'Observation de la Terre - SPOT 7
#   Julien, Ivan, Freddie (working name)     - JIF
# ------------------------------------------------------ #

SPOT_BANDS = [0, 1, 2, 3]
SPOT_RGB_BANDS = [1, 2, 3]
SPOT_MAX_EXPECTED_VALUE_8_BIT = 255
SPOT_MAX_EXPECTED_VALUE_12_BIT = 10000

ROOT_JIF_DATA = os.path.join("data", "JIF")
ROOT_JIF_DATA_TRAIN = os.path.join(ROOT_JIF_DATA, "train")

# Precomputed on training data by running: src/datasets.py
JIF_S2_MEAN = torch.Tensor(
    [
        0.0008,
        0.0117,
        0.0211,
        0.0198,
        0.0671,
        0.1274,
        0.1449,
        0.1841,
        0.1738,
        0.1803,
        0.0566,
        -0.0559,
    ]
).to(torch.float64)

JIF_S2_STD = torch.Tensor(
    [
        0.0892,
        0.0976,
        0.1001,
        0.1169,
        0.1154,
        0.1066,
        0.1097,
        0.1151,
        0.1122,
        0.1176,
        0.1298,
        0.1176,
    ]
).to(torch.float64)

URBAN_SPOT_MEAN = torch.Tensor([0.0025, 0.0022, 0.0017]).to(torch.float64)
URBAN_SPOT_STD = torch.Tensor([0.0021, 0.0018, 0.0017]).to(torch.float64)

	import os
	import torch
	from glob import glob

	# ------------------------------------------------------ #
	# Sentinel2 - S2
	# Level 2A - L2A
	# Level 1C - L1C
	# ------------------------------------------------------ #


	# Source: https://custom-scripts.sentinel-hub.com/custom-scripts/sentinel-2/composites/
	# NOTE: these are 1-indexed because GDAL is 1-indexed.
	S2_ALL_12BANDS = {
	"true_color": [4, 3, 2],
	"false_color": [8, 4, 3],
	"swir": [12, 8, 4],
	"agriculture": [11, 8, 2],
	"geology": [12, 11, 2],
	"bathimetric": [4, 3, 1],
	"true_color_zero_index": [3, 2, 1],
	}

	S2_ALL_BANDS = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]

	# ------------------------------------------------------ #
	# SpaceNet7 - SN7
	# PlanetScope - PS
	# ------------------------------------------------------ #

	PS_BANDS = [1, 2, 3]

	METADATA_PATH = 'dataset/metadata.csv'
	ROOT_SN7_DATA = os.path.join("data", "SN7")
	ROOT_SN7_DATA_TRAIN = os.path.join(ROOT_SN7_DATA, "train")
	SCENES_SN7 = glob(os.path.join(ROOT_SN7_DATA_TRAIN, "L15*"))

	SN7_SUBDIRECTORIES = {"lr": "S2", "lrc": "S2_masks", "hr": "images"}
	SN7_BANDS_TO_READ = {"lr": S2_ALL_12BANDS["true_color"], "lrc": None, "hr": PS_BANDS}


	# Precomputed on training data by running: src/datasets.py
	S2_SN7_MEAN = torch.Tensor(
	[
	0.0769,
	0.0906,
	0.1162,
	0.1318,
	0.1619,
	0.2275,
	0.2500,
	0.2588,
	0.2667,
	0.2684,
	0.2461,
	0.1874,
	]
	).to(torch.float64)

	S2_SN7_STD = torch.Tensor(
	[
	0.1708,
	0.1672,
	0.1614,
	0.1677,
	0.1698,
	0.1513,
	0.1490,
	0.1515,
	0.1461,
	0.2365,
	0.1472,
	0.1418,
	]
	).to(torch.float64)

	PS_MEAN = torch.Tensor([105.0, 92.0, 66.0]).to(torch.float64) # RGB
	PS_STD = torch.Tensor([58.2366, 47.1755, 44.9830]).to(torch.float64)
	SN7_MAX_EXPECTED_HR_VALUE = 255


	# ------------------------------------------------------ #
	# Satellite pour l'Observation de la Terre - SPOT 7
	# Julien, Ivan, Freddie (working name) - JIF
	# ------------------------------------------------------ #

	SPOT_BANDS = [0, 1, 2, 3]
	SPOT_RGB_BANDS = [1, 2, 3]
	SPOT_MAX_EXPECTED_VALUE_8_BIT = 255
	SPOT_MAX_EXPECTED_VALUE_12_BIT = 10000

	ROOT_JIF_DATA = os.path.join("data", "JIF")
	ROOT_JIF_DATA_TRAIN = os.path.join(ROOT_JIF_DATA, "train")

	# Precomputed on training data by running: src/datasets.py
	JIF_S2_MEAN = torch.Tensor(
	[
	0.0008,
	0.0117,
	0.0211,
	0.0198,
	0.0671,
	0.1274,
	0.1449,
	0.1841,
	0.1738,
	0.1803,
	0.0566,
	-0.0559,
	]
	).to(torch.float64)

	JIF_S2_STD = torch.Tensor(
	[
	0.0892,
	0.0976,
	0.1001,
	0.1169,
	0.1154,
	0.1066,
	0.1097,
	0.1151,
	0.1122,
	0.1176,
	0.1298,
	0.1176,
	]
	).to(torch.float64)

	URBAN_SPOT_MEAN = torch.Tensor([0.0025, 0.0022, 0.0017]).to(torch.float64)
	URBAN_SPOT_STD = torch.Tensor([0.0021, 0.0018, 0.0017]).to(torch.float64)