made utils folder
Browse files- GBI_16_4D.py → GBI-16-4D.py +2 -97
- README.md +1 -1
- utils/__init__.py +0 -0
- utils/create_splits.py +107 -0
GBI_16_4D.py → GBI-16-4D.py
RENAMED
@@ -47,7 +47,7 @@ _REPO_ID = "AstroCompress/GBI-16-4D"
|
|
47 |
class GBI_16_4D(datasets.GeneratorBasedBuilder):
|
48 |
"""GBI-16-4D Dataset"""
|
49 |
|
50 |
-
VERSION = datasets.Version("1.0.
|
51 |
|
52 |
BUILDER_CONFIGS = [
|
53 |
datasets.BuilderConfig(
|
@@ -136,99 +136,4 @@ class GBI_16_4D(datasets.GeneratorBasedBuilder):
|
|
136 |
task_instance_key = f"{self.config.name}-{split}-{idx}"
|
137 |
with fits.open(filepath, memmap=False, ignore_missing_simple=True) as hdul:
|
138 |
image_data = hdul[0].data.tolist()
|
139 |
-
yield task_instance_key, {**{"image": image_data}, **item}
|
140 |
-
|
141 |
-
|
142 |
-
def get_fits_footprint(fits_path):
|
143 |
-
"""
|
144 |
-
Process a FITS file to extract WCS information and calculate the footprint.
|
145 |
-
|
146 |
-
Parameters:
|
147 |
-
fits_path (str): Path to the FITS file.
|
148 |
-
|
149 |
-
Returns:
|
150 |
-
tuple: A tuple containing the WCS footprint coordinates.
|
151 |
-
"""
|
152 |
-
with fits.open(fits_path) as hdul:
|
153 |
-
hdul[0].data = hdul[0].data[0, 0]
|
154 |
-
wcs = WCS(hdul[0].header)
|
155 |
-
shape = sorted(tuple(wcs.pixel_shape))[:2]
|
156 |
-
footprint = wcs.calc_footprint(axes=shape)
|
157 |
-
coords = list(footprint.flatten())
|
158 |
-
return coords
|
159 |
-
|
160 |
-
def calculate_pixel_scale(header):
|
161 |
-
"""
|
162 |
-
Calculate the pixel scale in arcseconds per pixel from a FITS header.
|
163 |
-
|
164 |
-
Parameters:
|
165 |
-
header (astropy.io.fits.header.Header): The FITS header containing WCS information.
|
166 |
-
|
167 |
-
Returns:
|
168 |
-
Mean of the pixel scales in x and y.
|
169 |
-
"""
|
170 |
-
# Extract the CD matrix elements
|
171 |
-
cd1_1 = header.get('CD1_1', np.nan)
|
172 |
-
cd1_2 = header.get('CD1_2', np.nan)
|
173 |
-
cd2_1 = header.get('CD2_1', np.nan)
|
174 |
-
cd2_2 = header.get('CD2_2', np.nan)
|
175 |
-
|
176 |
-
# Calculate the pixel scales in arcseconds per pixel
|
177 |
-
pixscale_x = np.sqrt(cd1_1**2 + cd1_2**2) * 3600
|
178 |
-
pixscale_y = np.sqrt(cd2_1**2 + cd2_2**2) * 3600
|
179 |
-
|
180 |
-
return np.mean([pixscale_x, pixscale_y])
|
181 |
-
|
182 |
-
|
183 |
-
def make_split_jsonl_files(config_type="tiny", data_dir="./data",
|
184 |
-
outdir="./splits", seed=42):
|
185 |
-
"""
|
186 |
-
Create jsonl files for the GBI-16-4D dataset.
|
187 |
-
|
188 |
-
config_type: str, default="tiny"
|
189 |
-
The type of split to create. Options are "tiny" and "full".
|
190 |
-
data_dir: str, default="./data"
|
191 |
-
The directory where the FITS files are located.
|
192 |
-
outdir: str, default="./splits"
|
193 |
-
The directory where the jsonl files will be created.
|
194 |
-
seed: int, default=42
|
195 |
-
The seed for the random split.
|
196 |
-
"""
|
197 |
-
random.seed(seed)
|
198 |
-
os.makedirs(outdir, exist_ok=True)
|
199 |
-
|
200 |
-
fits_files = glob(os.path.join(data_dir, "*.fits"))
|
201 |
-
random.shuffle(fits_files)
|
202 |
-
if config_type == "tiny":
|
203 |
-
train_files = fits_files[:2]
|
204 |
-
test_files = fits_files[2:3]
|
205 |
-
elif config_type == "full":
|
206 |
-
split_idx = int(0.8 * len(fits_files))
|
207 |
-
train_files = fits_files[:split_idx]
|
208 |
-
test_files = fits_files[split_idx:]
|
209 |
-
else:
|
210 |
-
raise ValueError("Unsupported config_type. Use 'tiny' or 'full'.")
|
211 |
-
|
212 |
-
def create_jsonl(files, split_name):
|
213 |
-
output_file = os.path.join(outdir, f"{config_type}_{split_name}.jsonl")
|
214 |
-
with open(output_file, "w") as out_f:
|
215 |
-
for file in tqdm(files):
|
216 |
-
#print(file, flush=True, end="...")
|
217 |
-
with fits.open(file, memmap=False, ignore_missing_simple=True) as hdul:
|
218 |
-
image_id = os.path.basename(file).split(".fits")[0]
|
219 |
-
ra = hdul[0].header.get('CRVAL1', np.nan)
|
220 |
-
dec = hdul[0].header.get('CRVAL2', np.nan)
|
221 |
-
pixscale = calculate_pixel_scale(hdul[0].header)
|
222 |
-
ntimes = hdul[0].data.shape[0]
|
223 |
-
nbands = hdul[0].data.shape[1]
|
224 |
-
footprint = get_fits_footprint(file)
|
225 |
-
item = {"image_id": image_id, "image": file, "ra": ra, "dec": dec,
|
226 |
-
"pixscale": pixscale, "ntimes": ntimes, "nbands": nbands, "footprint": footprint}
|
227 |
-
out_f.write(json.dumps(item) + "\n")
|
228 |
-
|
229 |
-
create_jsonl(train_files, "train")
|
230 |
-
create_jsonl(test_files, "test")
|
231 |
-
|
232 |
-
if __name__ == "__main__":
|
233 |
-
make_split_jsonl_files("tiny")
|
234 |
-
make_split_jsonl_files("full")
|
|
|
47 |
class GBI_16_4D(datasets.GeneratorBasedBuilder):
|
48 |
"""GBI-16-4D Dataset"""
|
49 |
|
50 |
+
VERSION = datasets.Version("1.0.2")
|
51 |
|
52 |
BUILDER_CONFIGS = [
|
53 |
datasets.BuilderConfig(
|
|
|
136 |
task_instance_key = f"{self.config.name}-{split}-{idx}"
|
137 |
with fits.open(filepath, memmap=False, ignore_missing_simple=True) as hdul:
|
138 |
image_data = hdul[0].data.tolist()
|
139 |
+
yield task_instance_key, {**{"image": image_data}, **item}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
README.md
CHANGED
@@ -91,7 +91,7 @@ Then `cd GBI-16-4D` and start python like:
|
|
91 |
|
92 |
```python
|
93 |
from datasets import load_dataset
|
94 |
-
dataset = load_dataset("./
|
95 |
ds = dataset.with_format("np")
|
96 |
```
|
97 |
|
|
|
91 |
|
92 |
```python
|
93 |
from datasets import load_dataset
|
94 |
+
dataset = load_dataset("./GBI-16-4D.py", "tiny", data_dir="./data/")
|
95 |
ds = dataset.with_format("np")
|
96 |
```
|
97 |
|
utils/__init__.py
ADDED
File without changes
|
utils/create_splits.py
ADDED
@@ -0,0 +1,107 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
def get_fits_footprint(fits_path):
|
16 |
+
"""
|
17 |
+
Process a FITS file to extract WCS information and calculate the footprint.
|
18 |
+
|
19 |
+
Parameters:
|
20 |
+
fits_path (str): Path to the FITS file.
|
21 |
+
|
22 |
+
Returns:
|
23 |
+
tuple: A tuple containing the WCS footprint coordinates.
|
24 |
+
"""
|
25 |
+
with fits.open(fits_path) as hdul:
|
26 |
+
hdul[0].data = hdul[0].data[0, 0]
|
27 |
+
wcs = WCS(hdul[0].header)
|
28 |
+
shape = sorted(tuple(wcs.pixel_shape))[:2]
|
29 |
+
footprint = wcs.calc_footprint(axes=shape)
|
30 |
+
coords = list(footprint.flatten())
|
31 |
+
return coords
|
32 |
+
|
33 |
+
def calculate_pixel_scale(header):
|
34 |
+
"""
|
35 |
+
Calculate the pixel scale in arcseconds per pixel from a FITS header.
|
36 |
+
|
37 |
+
Parameters:
|
38 |
+
header (astropy.io.fits.header.Header): The FITS header containing WCS information.
|
39 |
+
|
40 |
+
Returns:
|
41 |
+
Mean of the pixel scales in x and y.
|
42 |
+
"""
|
43 |
+
# Extract the CD matrix elements
|
44 |
+
cd1_1 = header.get('CD1_1', np.nan)
|
45 |
+
cd1_2 = header.get('CD1_2', np.nan)
|
46 |
+
cd2_1 = header.get('CD2_1', np.nan)
|
47 |
+
cd2_2 = header.get('CD2_2', np.nan)
|
48 |
+
|
49 |
+
# Calculate the pixel scales in arcseconds per pixel
|
50 |
+
pixscale_x = np.sqrt(cd1_1**2 + cd1_2**2) * 3600
|
51 |
+
pixscale_y = np.sqrt(cd2_1**2 + cd2_2**2) * 3600
|
52 |
+
|
53 |
+
return np.mean([pixscale_x, pixscale_y])
|
54 |
+
|
55 |
+
|
56 |
+
def make_split_jsonl_files(config_type="tiny", data_dir="./data",
|
57 |
+
outdir="./splits", seed=42):
|
58 |
+
"""
|
59 |
+
Create jsonl files for the GBI-16-4D dataset.
|
60 |
+
|
61 |
+
config_type: str, default="tiny"
|
62 |
+
The type of split to create. Options are "tiny" and "full".
|
63 |
+
data_dir: str, default="./data"
|
64 |
+
The directory where the FITS files are located.
|
65 |
+
outdir: str, default="./splits"
|
66 |
+
The directory where the jsonl files will be created.
|
67 |
+
seed: int, default=42
|
68 |
+
The seed for the random split.
|
69 |
+
"""
|
70 |
+
random.seed(seed)
|
71 |
+
os.makedirs(outdir, exist_ok=True)
|
72 |
+
|
73 |
+
fits_files = glob(os.path.join(data_dir, "*.fits"))
|
74 |
+
random.shuffle(fits_files)
|
75 |
+
if config_type == "tiny":
|
76 |
+
train_files = fits_files[:2]
|
77 |
+
test_files = fits_files[2:3]
|
78 |
+
elif config_type == "full":
|
79 |
+
split_idx = int(0.8 * len(fits_files))
|
80 |
+
train_files = fits_files[:split_idx]
|
81 |
+
test_files = fits_files[split_idx:]
|
82 |
+
else:
|
83 |
+
raise ValueError("Unsupported config_type. Use 'tiny' or 'full'.")
|
84 |
+
|
85 |
+
def create_jsonl(files, split_name):
|
86 |
+
output_file = os.path.join(outdir, f"{config_type}_{split_name}.jsonl")
|
87 |
+
with open(output_file, "w") as out_f:
|
88 |
+
for file in tqdm(files):
|
89 |
+
#print(file, flush=True, end="...")
|
90 |
+
with fits.open(file, memmap=False, ignore_missing_simple=True) as hdul:
|
91 |
+
image_id = os.path.basename(file).split(".fits")[0]
|
92 |
+
ra = hdul[0].header.get('CRVAL1', np.nan)
|
93 |
+
dec = hdul[0].header.get('CRVAL2', np.nan)
|
94 |
+
pixscale = calculate_pixel_scale(hdul[0].header)
|
95 |
+
ntimes = hdul[0].data.shape[0]
|
96 |
+
nbands = hdul[0].data.shape[1]
|
97 |
+
footprint = get_fits_footprint(file)
|
98 |
+
item = {"image_id": image_id, "image": file, "ra": ra, "dec": dec,
|
99 |
+
"pixscale": pixscale, "ntimes": ntimes, "nbands": nbands, "footprint": footprint}
|
100 |
+
out_f.write(json.dumps(item) + "\n")
|
101 |
+
|
102 |
+
create_jsonl(train_files, "train")
|
103 |
+
create_jsonl(test_files, "test")
|
104 |
+
|
105 |
+
if __name__ == "__main__":
|
106 |
+
make_split_jsonl_files("tiny")
|
107 |
+
make_split_jsonl_files("full")
|