-
Notifications
You must be signed in to change notification settings - Fork 3
/
features_extraction.py
355 lines (279 loc) · 11.6 KB
/
features_extraction.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
import argparse
import logging
import os
from skimage import exposure
import numpy as np
import torch
from torchvision import transforms
from osgeo import gdal, gdal_array
import matplotlib.pyplot as plt
#from scipy import signal
import numpy
import cv2
from PIL import Image
from scipy.signal import fftconvolve
# import earthpy as et
# import earthpy.spatial as es
from unet import UNet_VAE
from unet import UNet_VAE_old, UNet_VAE_RQ_old, UNet_VAE_RQ_test, UNet_VAE_RQ_old_torch
from unet import UNet_VAE_RQ_new_torch, UNet_VAE_RQ_scheme3
from unet import UNet_VAE_RQ_scheme1, UNet_VAE_RQ_scheme2
from unet import UNet_test
from utils.utils import plot_img_and_mask, plot_img_and_mask_3, plot_img_and_mask_recon
#image_path = '/home/geoint/tri/github_files/test_img/number13458.TIF'
#mask_true_path = '/home/geoint/tri/github_files/test_label/number13458.TIF'
#image_path = '/home/geoint/tri/github_files/sentinel2_im/2016105_0.tif'
#mask_true_path = '/home/geoint/tri/github_files/sentinel2_im/2016105_0.tif'
#npy_path = '/home/geoint/tri/github_files/input_senegal/Tappan01_WV02_20110430_M1BS_103001000A27E100_data_568.npy'
file_path = '/home/geoint/tri/nasa_senegal/cassemance/Tappan02_WV02_20120218_M1BS_103001001077BE00_data.tif'
use_cuda = True
#device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
#im_type = image_path[30:38]
im_type='senegal'
segment=False
alpha = 0.1
unet_option = 'unet' # options: 'unet_vae_old','unet_vae_RQ_scheme1' 'unet_vae_RQ_scheme3'
image_option = "clean" # "clean" or "noisy"
##################################
def rescale(image): ## function to rescale image for visualization
map_img = np.zeros(image.shape)
for band in range(image.shape[2]):
p2, p98 = np.percentile(image[:,:,band], (2, 98))
map_img[:,:,band] = exposure.rescale_intensity(image[:,:,band], in_range=(p2, p98))
return map_img
def rescale_truncate(image): ## function to rescale image for visualization
if np.amin(image) < 0:
image = np.where(image < 0,0,image)
if np.amax(image) > 1:
image = np.where(image > 1,1,image)
map_img = np.zeros((256,256,3))
for band in range(3):
p2, p98 = np.percentile(image[:,:,band], (2, 98))
map_img[:,:,band] = exposure.rescale_intensity(image[:,:,band], in_range=(p2, p98))
return map_img
# def load_npy(file_path):
# data = np.load(file_path)
# data_1=(data - np.min(data)) / (np.max(data) - np.min(data))
# data_1 = rescale(data_1)
# # print(np.max(data))
# # print(np.min(data))
# # plt.imshow(data[:,:,:3])
# # plt.show()
# print("data shape: ", data_1.shape)
# row,col,ch= data_1.shape
# sigma = 0.01 ## choosing sigma based on the input images, 0.1-0.3 for NAIP images, 0.002 to 0.01 for sentinel2 images
# noisy = data_1 + sigma*np.random.randn(row,col,ch)
# transform_tensor = transforms.ToTensor()
# if use_cuda:
# noisy_tensor = transform_tensor(noisy).cuda()
# tensor = transform_tensor(data_1).cuda()
# return tensor.view([1]+list(tensor.shape)), noisy_tensor.view([1]+list(noisy_tensor.shape))
def jpg_to_tensor(filepath):
naip_fn = filepath
driverTiff = gdal.GetDriverByName('GTiff')
naip_ds = gdal.Open(naip_fn, 1)
nbands = naip_ds.RasterCount
# create an empty array, each column of the empty array will hold one band of data from the image
# loop through each band in the image nad add to the data array
data = np.empty((naip_ds.RasterXSize*naip_ds.RasterYSize, nbands))
for i in range(1, nbands+1):
band = naip_ds.GetRasterBand(i).ReadAsArray()
data[:, i-1] = band.flatten()
img_data = np.zeros((naip_ds.RasterYSize, naip_ds.RasterXSize, naip_ds.RasterCount),
gdal_array.GDALTypeCodeToNumericTypeCode(naip_ds.GetRasterBand(1).DataType))
for b in range(img_data.shape[2]):
img_data[:, :, b] = naip_ds.GetRasterBand(b + 1).ReadAsArray()
pil = np.array(img_data)
pil = pil.reshape((5000,5000,8))
#pil = pil/255
pil = pil[256:512,512:768, :]
print(pil.shape)
# add noise
row,col,ch= pil.shape
sigma = 0.08
noisy = pil + sigma*np.random.randn(row,col,ch)
#pil_to_tensor = torchvision.transforms.Compose([torchvision.transforms.ToTensor()])
transform_tensor = transforms.ToTensor()
if use_cuda:
noisy_tensor = transform_tensor(noisy).cuda()
tensor = transform_tensor(pil).cuda()
return tensor.view([1]+list(tensor.shape)), noisy_tensor.view([1]+list(noisy_tensor.shape))
#accept a torch tensor, convert it to a jpg at a certain path
def tensor_to_jpg(tensor):
tensor = tensor.squeeze(0)
if use_cuda:
tensor = tensor.cpu()
pil = tensor.permute(1, 2, 0).numpy()
pil = np.array(pil)
pil = rescale(pil)
#pil = rescale_truncate(pil)
return pil
#predict image
def extract_features(net,
filepath,
device,
scale_factor=1,
out_threshold=0.5):
net.eval()
#img = img.unsqueeze(0)
if image_option=='clean':
img = jpg_to_tensor(filepath)[0] ## clean image
elif image_option=='noisy':
img = jpg_to_tensor(filepath)[1] ## noisy image
img = img.to(device=device, dtype=torch.float32)
print("input image shape: ", img.shape)
##### FEATURE EXTRACTION LOOP
# placeholders
PREDS = []
FEATS = []
# placeholder for batch features
features = {}
##### REGISTER HOOK
net.down_convs[1].pool.register_forward_hook(get_features(features, 'feats'))
# forward pass [with feature extraction]
preds = net(img)
if unet_option == 'unet':
preds = preds
else:
preds = preds[3]
# add feats and preds to lists
PREDS.append(preds.detach().cpu().numpy())
FEATS.append(features['feats'].cpu().numpy())
##### INSPECT FEATURES
PREDS = np.concatenate(PREDS)
FEATS = np.concatenate(FEATS)
print('- preds shape:', PREDS.shape)
print('- feats shape:', FEATS.shape)
return preds.detach().cpu(), FEATS, img.detach().cpu()
##### HELPER FUNCTION FOR FEATURE EXTRACTION
def get_features(features, name):
def hook(model, input, output):
features[name] = output.detach()
return hook
def get_false_color(image):
false_color = np.zeros((image.shape[0], image.shape[1], 3))
false_color[:,:,0] = image[:,:,6]
false_color[:,:,1] = image[:,:,2]
false_color[:,:,2] = image[:,:,1]
return false_color
# deal with nan
def nan_ptp(a):
return np.ptp(a[np.isfinite(a)])
## Function to calculate correlation
def ssim(im1, im2, window, k=(0.01, 0.03), l=255):
"""See https://ece.uwaterloo.ca/~z70wang/research/ssim/"""
# Check if the window is smaller than the images.
for a, b in zip(window.shape, im1.shape):
if a > b:
return None, None
# Values in k must be positive according to the base implementation.
for ki in k:
if ki < 0:
return None, None
c1 = (k[0] * l) ** 2
c2 = (k[1] * l) ** 2
window = window/numpy.sum(window)
mu1 = fftconvolve(im1, window, mode='valid')
mu2 = fftconvolve(im2, window, mode='valid')
mu1_sq = mu1 * mu1
mu2_sq = mu2 * mu2
mu1_mu2 = mu1 * mu2
sigma1_sq = fftconvolve(im1 * im1, window, mode='valid') - mu1_sq
sigma2_sq = fftconvolve(im2 * im2, window, mode='valid') - mu2_sq
sigma12 = fftconvolve(im1 * im2, window, mode='valid') - mu1_mu2
if c1 > 0 and c2 > 0:
num = (2 * mu1_mu2 + c1) * (2 * sigma12 + c2)
den = (mu1_sq + mu2_sq + c1) * (sigma1_sq + sigma2_sq + c2)
ssim_map = num / den
else:
num1 = 2 * mu1_mu2 + c1
num2 = 2 * sigma12 + c2
den1 = mu1_sq + mu2_sq + c1
den2 = sigma1_sq + sigma2_sq + c2
ssim_map = numpy.ones(numpy.shape(mu1))
index = (den1 * den2) > 0
ssim_map[index] = (num1[index] * num2[index]) / (den1[index] * den2[index])
index = (den1 != 0) & (den2 == 0)
ssim_map[index] = num1[index] / den1[index]
mssim = ssim_map.mean()
return mssim, ssim_map
def nrmse(im1, im2):
a, b = im1.shape
rmse = numpy.sqrt(numpy.sum((im2 - im1) ** 2) / float(a * b))
max_val = max(numpy.max(im1), numpy.max(im2))
min_val = min(numpy.min(im1), numpy.min(im2))
return 1 - (rmse / (max_val - min_val))
if __name__ == '__main__':
if unet_option == 'unet_vae_1':
net = UNet_VAE(8)
elif unet_option == 'unet':
net = UNet_test(8, in_channels=8)
elif unet_option == 'unet_vae_old':
net = UNet_VAE_old(8, segment, in_channels=8)
elif unet_option == 'unet_vae_RQ_old':
net = UNet_VAE_RQ_old(8, alpha)
# elif unet_option == 'unet_vae_RQ_allskip_trainable':
# net = UNet_VAE_RQ_old_trainable(8,alpha)
elif unet_option == 'unet_vae_RQ_torch':
net = UNet_VAE_RQ_new_torch(8, segment, alpha)
elif unet_option == 'unet_vae_RQ_scheme3':
net = UNet_VAE_RQ_scheme3(8, segment, alpha)
elif unet_option == 'unet_vae_RQ_scheme1':
net = UNet_VAE_RQ_scheme1(8, segment, alpha, in_channels=8)
elif unet_option == 'unet_vae_RQ_scheme2':
net = UNet_VAE_RQ_scheme2(8, segment, alpha)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
#logging.info(f'Loading model {args.model}')
logging.info(f'Using device {device}')
# model_saved = '/home/geoint/tri/github_files/github_checkpoints/checkpoint_unet_vae_old_4-18_epoch10_0.0_recon.pth'
# model_sentinel_saved = '/home/geoint/tri/github_files/github_checkpoints/checkpoint_unet_vae_old_epoch20_sentinel_4-28_recon.pth'
net.to(device=device)
#net.load_state_dict(torch.load(model_saved, map_location=device))
#print(net.down_convs[0])
#print(net.down_convs[0].pool)
logging.info('Model loaded!')
#logging.info(f'\nPredicting image {image_path} ...')
pred, feats, im = extract_features(net=net,
filepath=file_path,
scale_factor=1,
out_threshold=0.5,
device=device)
im = tensor_to_jpg(im)
#print(im.shape)
im = im.reshape((256,256,8))
pred = tensor_to_jpg(pred)
im_false = get_false_color(im)
pred_false = get_false_color(pred)
# calculate ndvi
ndvi_upper = im[:,:,6]-im[:,:,0]
ndvi_upper = np.array(ndvi_upper, dtype=float)
ndvi_lower = im[:,:,6]+im[:,:,0]
ndvi_lower = np.array(ndvi_lower, dtype=float)
ndvi = np.divide(ndvi_upper, ndvi_lower, out=np.zeros_like(ndvi_upper, dtype=float), where=ndvi_lower!=0)
feats = feats.reshape((feats.shape[1],feats.shape[2],feats.shape[3]))
print('feats shape: ', feats.shape)
h = feats.shape[1]
w = feats.shape[2]
ndvi_h, ndvi_w = ndvi.shape
bin_size = ndvi_h // h
ndvi_res = ndvi.reshape((h, bin_size,
h, bin_size, 1)).max(3).max(1)
print('ndvi_res max: ', np.max(ndvi_res))
print('ndvi_res min: ', np.min(ndvi_res))
b = np.zeros((feats.shape[1],feats.shape[2]))
for i in range(feats.shape[0]):
#plot_img_and_mask_recon(im_false, feats[:,i,:,:].reshape((feats.shape[2],feats.shape[3])))
a = feats[i,:,:].reshape((feats.shape[1],feats.shape[2]))
h,w = a.shape
a = 2.*(a - np.min(a))/nan_ptp(a)-1
# ndvi_res = np.array(ndvi_res)
# print('ndvi type: ', type(ndvi_res))
# print('feat max: ', np.max(a))
# print('feat min: ', np.min(a))
nrmse_val = nrmse(a, ndvi_res)
if nrmse_val > -2:
#print(nrmse_val)
b += a
plot_img_and_mask_recon(ndvi_res, b)
plot_img_and_mask_recon(im_false, pred_false)
plot_img_and_mask_recon(im[:,:,:3], pred[:,:,:3])