utils folder
Browse files- SBI-16-3D.py +1 -93
- utils/__init__.py +0 -0
- utils/create_splits.py +105 -0
SBI-16-3D.py
CHANGED
@@ -129,96 +129,4 @@ class SBI_16_4D(datasets.GeneratorBasedBuilder):
|
|
129 |
# the first axis is integrations one, so we take the first element
|
130 |
# the second axis is the groups (time) axis and varies between images
|
131 |
image_data = hdul["SCI"].data[0,:,:,:].tolist()
|
132 |
-
yield task_instance_key, {**{"image": image_data}, **item}
|
133 |
-
|
134 |
-
|
135 |
-
def get_fits_footprint(fits_path):
|
136 |
-
"""
|
137 |
-
Process a FITS file to extract WCS information and calculate the footprint.
|
138 |
-
|
139 |
-
Parameters:
|
140 |
-
fits_path (str): Path to the FITS file.
|
141 |
-
|
142 |
-
Returns:
|
143 |
-
tuple: A tuple containing the WCS footprint coordinates.
|
144 |
-
"""
|
145 |
-
with fits.open(fits_path) as hdul:
|
146 |
-
hdul[1].data = hdul[1].data[0, 0]
|
147 |
-
wcs = WCS(hdul[1].header)
|
148 |
-
shape = sorted(tuple(wcs.pixel_shape))[:2]
|
149 |
-
footprint = wcs.calc_footprint(axes=shape)
|
150 |
-
coords = list(footprint.flatten())
|
151 |
-
return coords
|
152 |
-
|
153 |
-
|
154 |
-
def calculate_pixel_scale(header):
|
155 |
-
"""
|
156 |
-
Calculate the pixel scale in arcseconds per pixel from a FITS header.
|
157 |
-
|
158 |
-
Parameters:
|
159 |
-
header (astropy.io.fits.header.Header): The FITS header containing WCS information.
|
160 |
-
|
161 |
-
Returns:
|
162 |
-
Mean of the pixel scales in x and y.
|
163 |
-
"""
|
164 |
-
|
165 |
-
# Calculate the pixel scales in arcseconds per pixel
|
166 |
-
pixscale_x = header.get('CDELT1', np.nan)
|
167 |
-
pixscale_y = header.get('CDELT2', np.nan)
|
168 |
-
|
169 |
-
return np.mean([pixscale_x, pixscale_y])
|
170 |
-
|
171 |
-
|
172 |
-
def make_split_jsonl_files(config_type="tiny", data_dir="./data",
|
173 |
-
outdir="./splits", seed=42):
|
174 |
-
"""
|
175 |
-
Create jsonl files for the SBI-16-3D dataset.
|
176 |
-
|
177 |
-
config_type: str, default="tiny"
|
178 |
-
The type of split to create. Options are "tiny" and "full".
|
179 |
-
data_dir: str, default="./data"
|
180 |
-
The directory where the FITS files are located.
|
181 |
-
outdir: str, default="./splits"
|
182 |
-
The directory where the jsonl files will be created.
|
183 |
-
seed: int, default=42
|
184 |
-
The seed for the random split.
|
185 |
-
"""
|
186 |
-
random.seed(seed)
|
187 |
-
os.makedirs(outdir, exist_ok=True)
|
188 |
-
|
189 |
-
fits_files = glob(os.path.join(data_dir, "*.fits"))
|
190 |
-
random.shuffle(fits_files)
|
191 |
-
if config_type == "tiny":
|
192 |
-
train_files = fits_files[:2]
|
193 |
-
test_files = fits_files[2:3]
|
194 |
-
elif config_type == "full":
|
195 |
-
split_idx = int(0.8 * len(fits_files))
|
196 |
-
train_files = fits_files[:split_idx]
|
197 |
-
test_files = fits_files[split_idx:]
|
198 |
-
else:
|
199 |
-
raise ValueError("Unsupported config_type. Use 'tiny' or 'full'.")
|
200 |
-
|
201 |
-
def create_jsonl(files, split_name):
|
202 |
-
output_file = os.path.join(outdir, f"{config_type}_{split_name}.jsonl")
|
203 |
-
with open(output_file, "w") as out_f:
|
204 |
-
for file in tqdm(files):
|
205 |
-
#print(file, flush=True, end="...")
|
206 |
-
with fits.open(file, memmap=False) as hdul:
|
207 |
-
image_id = os.path.basename(file).split(".fits")[0]
|
208 |
-
ra = hdul["SCI"].header.get('CRVAL1', 0)
|
209 |
-
dec = hdul["SCI"].header.get('CRVAL2', 0)
|
210 |
-
pixscale = calculate_pixel_scale(hdul["SCI"].header)
|
211 |
-
footprint = get_fits_footprint(file)
|
212 |
-
# get the number of groups per int
|
213 |
-
ntimes = hdul["SCI"].data.shape[1]
|
214 |
-
item = {"image_id": image_id, "image": file, "ra": ra, "dec": dec,
|
215 |
-
"pixscale": pixscale, "ntimes": ntimes, "footprint": footprint}
|
216 |
-
out_f.write(json.dumps(item) + "\n")
|
217 |
-
|
218 |
-
create_jsonl(train_files, "train")
|
219 |
-
create_jsonl(test_files, "test")
|
220 |
-
|
221 |
-
|
222 |
-
if __name__ == "__main__":
|
223 |
-
make_split_jsonl_files("tiny")
|
224 |
-
make_split_jsonl_files("full")
|
|
|
129 |
# the first axis is integrations one, so we take the first element
|
130 |
# the second axis is the groups (time) axis and varies between images
|
131 |
image_data = hdul["SCI"].data[0,:,:,:].tolist()
|
132 |
+
yield task_instance_key, {**{"image": image_data}, **item}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
utils/__init__.py
ADDED
File without changes
|
utils/create_splits.py
ADDED
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import random
|
3 |
+
from glob import glob
|
4 |
+
import json
|
5 |
+
from huggingface_hub import hf_hub_download
|
6 |
+
from tqdm import tqdm
|
7 |
+
import numpy as np
|
8 |
+
|
9 |
+
from astropy.io import fits
|
10 |
+
from astropy.wcs import WCS
|
11 |
+
import datasets
|
12 |
+
from datasets import DownloadManager
|
13 |
+
from fsspec.core import url_to_fs
|
14 |
+
|
15 |
+
|
16 |
+
def get_fits_footprint(fits_path):
|
17 |
+
"""
|
18 |
+
Process a FITS file to extract WCS information and calculate the footprint.
|
19 |
+
|
20 |
+
Parameters:
|
21 |
+
fits_path (str): Path to the FITS file.
|
22 |
+
|
23 |
+
Returns:
|
24 |
+
tuple: A tuple containing the WCS footprint coordinates.
|
25 |
+
"""
|
26 |
+
with fits.open(fits_path) as hdul:
|
27 |
+
hdul[1].data = hdul[1].data[0, 0]
|
28 |
+
wcs = WCS(hdul[1].header)
|
29 |
+
shape = sorted(tuple(wcs.pixel_shape))[:2]
|
30 |
+
footprint = wcs.calc_footprint(axes=shape)
|
31 |
+
coords = list(footprint.flatten())
|
32 |
+
return coords
|
33 |
+
|
34 |
+
|
35 |
+
def calculate_pixel_scale(header):
|
36 |
+
"""
|
37 |
+
Calculate the pixel scale in arcseconds per pixel from a FITS header.
|
38 |
+
|
39 |
+
Parameters:
|
40 |
+
header (astropy.io.fits.header.Header): The FITS header containing WCS information.
|
41 |
+
|
42 |
+
Returns:
|
43 |
+
Mean of the pixel scales in x and y.
|
44 |
+
"""
|
45 |
+
|
46 |
+
# Calculate the pixel scales in arcseconds per pixel
|
47 |
+
pixscale_x = header.get('CDELT1', np.nan)
|
48 |
+
pixscale_y = header.get('CDELT2', np.nan)
|
49 |
+
|
50 |
+
return np.mean([pixscale_x, pixscale_y])
|
51 |
+
|
52 |
+
|
53 |
+
def make_split_jsonl_files(config_type="tiny", data_dir="./data",
|
54 |
+
outdir="./splits", seed=42):
|
55 |
+
"""
|
56 |
+
Create jsonl files for the SBI-16-3D dataset.
|
57 |
+
|
58 |
+
config_type: str, default="tiny"
|
59 |
+
The type of split to create. Options are "tiny" and "full".
|
60 |
+
data_dir: str, default="./data"
|
61 |
+
The directory where the FITS files are located.
|
62 |
+
outdir: str, default="./splits"
|
63 |
+
The directory where the jsonl files will be created.
|
64 |
+
seed: int, default=42
|
65 |
+
The seed for the random split.
|
66 |
+
"""
|
67 |
+
random.seed(seed)
|
68 |
+
os.makedirs(outdir, exist_ok=True)
|
69 |
+
|
70 |
+
fits_files = glob(os.path.join(data_dir, "*.fits"))
|
71 |
+
random.shuffle(fits_files)
|
72 |
+
if config_type == "tiny":
|
73 |
+
train_files = fits_files[:2]
|
74 |
+
test_files = fits_files[2:3]
|
75 |
+
elif config_type == "full":
|
76 |
+
split_idx = int(0.8 * len(fits_files))
|
77 |
+
train_files = fits_files[:split_idx]
|
78 |
+
test_files = fits_files[split_idx:]
|
79 |
+
else:
|
80 |
+
raise ValueError("Unsupported config_type. Use 'tiny' or 'full'.")
|
81 |
+
|
82 |
+
def create_jsonl(files, split_name):
|
83 |
+
output_file = os.path.join(outdir, f"{config_type}_{split_name}.jsonl")
|
84 |
+
with open(output_file, "w") as out_f:
|
85 |
+
for file in tqdm(files):
|
86 |
+
#print(file, flush=True, end="...")
|
87 |
+
with fits.open(file, memmap=False) as hdul:
|
88 |
+
image_id = os.path.basename(file).split(".fits")[0]
|
89 |
+
ra = hdul["SCI"].header.get('CRVAL1', 0)
|
90 |
+
dec = hdul["SCI"].header.get('CRVAL2', 0)
|
91 |
+
pixscale = calculate_pixel_scale(hdul["SCI"].header)
|
92 |
+
footprint = get_fits_footprint(file)
|
93 |
+
# get the number of groups per int
|
94 |
+
ntimes = hdul["SCI"].data.shape[1]
|
95 |
+
item = {"image_id": image_id, "image": file, "ra": ra, "dec": dec,
|
96 |
+
"pixscale": pixscale, "ntimes": ntimes, "footprint": footprint}
|
97 |
+
out_f.write(json.dumps(item) + "\n")
|
98 |
+
|
99 |
+
create_jsonl(train_files, "train")
|
100 |
+
create_jsonl(test_files, "test")
|
101 |
+
|
102 |
+
|
103 |
+
if __name__ == "__main__":
|
104 |
+
make_split_jsonl_files("tiny")
|
105 |
+
make_split_jsonl_files("full")
|