wangyi111 commited on
Commit
c9dfc89
·
verified ·
1 Parent(s): a9456dc

Upload dataset_biomass_s3olci.py

Browse files
biomass_s3olci/dataset_biomass_s3olci.py CHANGED
@@ -4,6 +4,8 @@ import os
4
  import rasterio
5
  import torch
6
  import numpy as np
 
 
7
 
8
  S3_OLCI_SCALE = [0.0139465,0.0133873,0.0121481,0.0115198,0.0100953,0.0123538,0.00879161,0.00876539,
9
  0.0095103,0.00773378,0.00675523,0.0071996,0.00749684,0.0086512,0.00526779,0.00530267,
@@ -16,7 +18,7 @@ BIOMASS_STD = 110.5369
16
  class S3OLCI_BiomassDataset(Dataset):
17
  '''
18
  4000/1000 train/test images 94x94x21 (full dataset is 25K)
19
- CCI biomass 282x282
20
  nodata: -inf
21
  time series: 1-4 images / location
22
 
@@ -32,6 +34,18 @@ class S3OLCI_BiomassDataset(Dataset):
32
 
33
  self.fnames = os.listdir(self.biomass_dir)
34
 
 
 
 
 
 
 
 
 
 
 
 
 
35
  def __len__(self):
36
  return len(self.fnames)
37
 
@@ -39,13 +53,19 @@ class S3OLCI_BiomassDataset(Dataset):
39
  fname = self.fnames[idx]
40
  biomass_path = os.path.join(self.biomass_dir, fname)
41
  s3_path = os.path.join(self.img_dir, fname.replace('.tif',''))
42
- img_fnames = os.listdir(s3_path)
43
- s3_paths = []
44
- for img_fname in img_fnames:
45
- s3_paths.append(os.path.join(s3_path, img_fname))
46
 
 
 
 
 
 
 
 
 
 
47
  imgs = []
48
  img_paths = []
 
49
  for img_path in s3_paths:
50
  with rasterio.open(img_path) as src:
51
  img = src.read()
@@ -59,12 +79,30 @@ class S3OLCI_BiomassDataset(Dataset):
59
  for b in range(21):
60
  img[b] = img[b]*S3_OLCI_SCALE[b]
61
  img = torch.from_numpy(img).float()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  imgs.append(img)
63
  img_paths.append(img_path)
64
- # pad to 4 images if less than 4
65
- while len(imgs) < 4:
66
- imgs.append(img)
67
- img_paths.append(img_path)
 
 
 
68
 
69
  with rasterio.open(biomass_path) as src:
70
  biomass = src.read(1)
@@ -72,6 +110,6 @@ class S3OLCI_BiomassDataset(Dataset):
72
  biomass = torch.from_numpy(biomass.astype('float32'))
73
  biomass = (biomass - BIOMASS_MEAN) / BIOMASS_STD # 0-center normalized
74
  if self.mode == 'static':
75
- return imgs[0], biomass # 94x94x21, 282x282
76
  elif self.mode == 'series':
77
- return imgs[0], imgs[1], imgs[2], imgs[3], biomass # 94x94x21, 94x94x21, 94x94x21, 94x94x21, 282x282
 
4
  import rasterio
5
  import torch
6
  import numpy as np
7
+ from pyproj import Transformer
8
+ from datetime import date
9
 
10
  S3_OLCI_SCALE = [0.0139465,0.0133873,0.0121481,0.0115198,0.0100953,0.0123538,0.00879161,0.00876539,
11
  0.0095103,0.00773378,0.00675523,0.0071996,0.00749684,0.0086512,0.00526779,0.00530267,
 
18
  class S3OLCI_BiomassDataset(Dataset):
19
  '''
20
  4000/1000 train/test images 94x94x21 (full dataset is 25K)
21
+ CCI biomass regression 282x282
22
  nodata: -inf
23
  time series: 1-4 images / location
24
 
 
34
 
35
  self.fnames = os.listdir(self.biomass_dir)
36
 
37
+ if self.mode == 'static':
38
+ self.static_csv = os.path.join(root_dir, split, 'static_fnames.csv')
39
+ with open(self.static_csv, 'r') as f:
40
+ lines = f.readlines()
41
+ self.static_img = {}
42
+ for line in lines:
43
+ dirname = line.strip().split(',')[0]
44
+ img_fname = line.strip().split(',')[1]
45
+ self.static_img[dirname] = img_fname
46
+
47
+
48
+
49
  def __len__(self):
50
  return len(self.fnames)
51
 
 
53
  fname = self.fnames[idx]
54
  biomass_path = os.path.join(self.biomass_dir, fname)
55
  s3_path = os.path.join(self.img_dir, fname.replace('.tif',''))
 
 
 
 
56
 
57
+ if self.mode == 'static':
58
+ img_fname = self.static_img[fname.replace('.tif','')]
59
+ s3_paths = [os.path.join(s3_path, img_fname)]
60
+ else:
61
+ img_fnames = os.listdir(s3_path)
62
+ s3_paths = []
63
+ for img_fname in img_fnames:
64
+ s3_paths.append(os.path.join(s3_path, img_fname))
65
+
66
  imgs = []
67
  img_paths = []
68
+ meta_infos = []
69
  for img_path in s3_paths:
70
  with rasterio.open(img_path) as src:
71
  img = src.read()
 
79
  for b in range(21):
80
  img[b] = img[b]*S3_OLCI_SCALE[b]
81
  img = torch.from_numpy(img).float()
82
+
83
+ if self.meta:
84
+ cx,cy = src.xy(src.height // 2, src.width // 2)
85
+ #crs_transformer = Transformer.from_crs(src.crs, 'epsg:4326')
86
+ #lon, lat = crs_transformer.transform(cx,cy)
87
+ lon, lat = cx, cy
88
+ img_fname = os.path.basename(img_path)
89
+ date_str = img_fname.split('_')[1][:8]
90
+ date_obj = date(int(date_str[:4]), int(date_str[4:6]), int(date_str[6:8]))
91
+ delta = (date_obj - self.reference_date).days
92
+ meta_info = np.array([lon, lat, delta, np.nan]).astype(np.float32)
93
+ else:
94
+ meta_info = np.array([np.nan,np.nan,np.nan,np.nan]).astype(np.float32)
95
+
96
+
97
  imgs.append(img)
98
  img_paths.append(img_path)
99
+
100
+ if self.mode == 'series':
101
+ # pad to 4 images if less than 4
102
+ while len(imgs) < 4:
103
+ imgs.append(img)
104
+ img_paths.append(img_path)
105
+ meta_infos.append(meta_info)
106
 
107
  with rasterio.open(biomass_path) as src:
108
  biomass = src.read(1)
 
110
  biomass = torch.from_numpy(biomass.astype('float32'))
111
  biomass = (biomass - BIOMASS_MEAN) / BIOMASS_STD # 0-center normalized
112
  if self.mode == 'static':
113
+ return imgs[0], meta_infos[0], biomass # 94x94x21, 282x282
114
  elif self.mode == 'series':
115
+ return imgs[0], imgs[1], imgs[2], imgs[3], meta_infos[0], meta_infos[1], meta_infos[2], meta_infos[3], biomass # 94x94x21, 94x94x21, 94x94x21, 94x94x21, 282x282