hayden-donnelly commited on
Commit
a7d0de2
1 Parent(s): 09d61a5

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +111 -0
README.md ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ task_categories:
4
+ - unconditional-image-generation
5
+ - image-classification
6
+ - text-to-image
7
+ pretty_name: World Heightmaps 360 V1
8
+ size_categories:
9
+ - 100K<n<1M
10
+ ---
11
+ # World Heightmaps 360px
12
+ This is a dataset of 360x360 Earth heightmaps generated from [SRTM 1 Arc-Second Global](https://huggingface.co/datasets/hayden-donnelly/srtm-1-arc-second-global).
13
+ Each heightmap is labelled according to its latitude and longitude. There are 573,995 samples.
14
+
15
+ ## Method
16
+ 1. Convert GeoTIFFs into PNGs with Python and Rasterio.
17
+ ```python
18
+ import rasterio
19
+ import matplotlib.pyplot as plt
20
+ import os
21
+
22
+ input_directory = '...'
23
+ output_directory = '...'
24
+ file_list = os.listdir(input_directory)
25
+
26
+ for i in range(len(file_list)):
27
+ image = rasterio.open(input_directory + file_list[i])
28
+ plt.imsave(output_directory + file_list[i][0:-4] + '.png', image.read(1), cmap='gray')
29
+ ```
30
+
31
+ 2. Split PNGs into 100 patches with Split Image.
32
+ ```python
33
+ from split_image import split_image
34
+ import os
35
+
36
+ input_directory = '...'
37
+ output_directory = '...'
38
+ file_list = os.listdir(input_directory)
39
+
40
+ for i in range(len(file_list)):
41
+ split_image(input_directory + file_list[i], 10, 10, should_square=True, should_cleanup=False, output_dir=output_directory)
42
+ ```
43
+
44
+ 3. Hand pick a dataset of corrupted and uncorrupted heightmaps then train a discriminator to automatically filter the whole dataset.
45
+
46
+ 4. Compile images into parquet files.
47
+ ```python
48
+ import pyarrow as pa
49
+ import pyarrow.parquet as pq
50
+ import pandas as pd
51
+ from PIL import Image
52
+ import os
53
+ import io
54
+ import json
55
+
56
+ samples_per_file = 6_000
57
+
58
+ root_dir = 'data/datasets/world-heightmaps-360px-png'
59
+ df = pd.read_csv(os.path.join(root_dir, 'metadata.csv'))
60
+ df = df.sample(frac=1).reset_index(drop=True)
61
+
62
+ def save_table(image_data, table_number):
63
+ print(f'Entries in table {table_number}: {len(image_data)}')
64
+ schema = pa.schema(
65
+ fields=[
66
+ ('heightmap', pa.struct([('bytes', pa.binary()), ('path', pa.string())])),
67
+ ('latitude', pa.string()),
68
+ ('longitude', pa.string())
69
+ ],
70
+ metadata={
71
+ b'huggingface': json.dumps({
72
+ 'info': {
73
+ 'features': {
74
+ 'heightmap': {'_type': 'Image'},
75
+ 'latitude': {'_type': 'Value', 'dtype': 'string'},
76
+ 'longitude': {'_type': 'Value', 'dtype': 'string'}
77
+ }
78
+ }
79
+ }).encode('utf-8')
80
+ }
81
+ )
82
+
83
+ table = pa.Table.from_pylist(image_data, schema=schema)
84
+ pq.write_table(table, f'data/world-heightmaps-360px-parquet/{str(table_number).zfill(4)}.parquet')
85
+
86
+ image_data = []
87
+ samples_in_current_file = 0
88
+ current_file_number = 0
89
+ for i, row in df.iterrows():
90
+ if samples_in_current_file >= samples_per_file:
91
+ save_table(image_data, current_file_number)
92
+ image_data = []
93
+ samples_in_current_file = 0
94
+ current_file_number += 1
95
+ samples_in_current_file += 1
96
+ image_path = row['file_name']
97
+ with Image.open(os.path.join(root_dir, image_path)) as image:
98
+ image_bytes = io.BytesIO()
99
+ image.save(image_bytes, format='PNG')
100
+ image_dict = {
101
+ 'heightmap': {
102
+ 'bytes': image_bytes.getvalue(),
103
+ 'path': image_path
104
+ },
105
+ 'latitude': str(row['latitude']),
106
+ 'longitude': str(row['longitude'])
107
+ }
108
+ image_data.append(image_dict)
109
+
110
+ save_table(image_data, current_file_number)
111
+ ```