-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpredict.py
185 lines (152 loc) · 7.58 KB
/
predict.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
import argparse
import pathlib
import importlib
from conf import default, general, paths
import os
import time
import sys
from utils.dataloader import PredDataSet
from torch.utils.data import DataLoader
import torch
from tqdm import tqdm
import numpy as np
from utils.ops import save_geotiff
from multiprocessing import Process, freeze_support
from torch.multiprocessing import freeze_support
import logging
parser = argparse.ArgumentParser(
description='Train NUMBER_MODELS models based in the same parameters'
)
parser.add_argument( # Experiment number
'-e', '--experiment',
type = int,
default = 1,
help = 'The number of the experiment'
)
parser.add_argument( # batch size
'-b', '--batch-size',
type = int,
default = default.PREDICTION_BATCH_SIZE,
help = 'The number of samples of each batch'
)
parser.add_argument( # Number of models to be trained
'-n', '--number-models',
type = int,
default = default.N_MODELS,
help = 'The number models to be trained from the scratch'
)
parser.add_argument( # Experiment path
'-x', '--experiments-path',
type = pathlib.Path,
default = paths.EXPERIMENTS_PATH,
help = 'The patch to data generated by all experiments'
)
parser.add_argument( # Reference year to predict
'-y', '--year',
type = int,
default = default.PROCESSED_YEARS[2],
help = 'Reference year to predict'
)
parser.add_argument( # base image
'-i', '--base-image-path',
type = pathlib.Path,
default = paths.OPT_PATH,
help = 'Path to the folder of the geotiff files to generate aligned labels of the same region'
)
args = parser.parse_args()
exp_path = os.path.join(str(args.experiments_path), f'exp_{args.experiment}')
logs_path = os.path.join(exp_path, f'logs')
models_path = os.path.join(exp_path, f'models')
visual_path = os.path.join(exp_path, f'visual')
predicted_path = os.path.join(exp_path, f'predicted')
results_path = os.path.join(exp_path, f'results')
device = "cuda" if torch.cuda.is_available() else "cpu"
#print(f"Using {device} device")
#def run(model_idx):
#if __name__ == '__main__':
# freeze_support()
# for model_idx in tqdm(range(args.number_models), desc='Model Idx'):
def run(model_idx):
outfile = os.path.join(logs_path, f'pred_{args.experiment}_{model_idx}.txt')
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s:%(levelname)s:%(name)s:%(message)s',
filename=outfile,
filemode='w'
)
log = logging.getLogger('predict')
model_m =importlib.import_module(f'conf.exp_{args.experiment}')
model = model_m.get_model(log)
model.to(device)
log.info(model)
model_path = os.path.join(models_path, f'model_{model_idx}.pth')
model.load_state_dict(torch.load(model_path))
torch.set_num_threads(8)
overlaps = general.PREDICTION_OVERLAPS
log.info(f'Overlaps pred: {overlaps}')
one_window = np.ones((general.PATCH_SIZE, general.PATCH_SIZE, general.N_CLASSES))
total_time = 0
n_images = 0
for im_0 in tqdm(range(general.N_IMAGES_YEAR), leave=False, desc='Img 0'):
for im_1 in tqdm(range(general.N_IMAGES_YEAR), leave = False, desc='Img 1'):
img_pair = (im_0, im_1)
dataset = PredDataSet(device = device, year = args.year, img_pair = img_pair)
label = dataset.label
log.info(f'Optical Image Year 0:{dataset.opt_file_0}')
log.info(f'Optical Image Year 1:{dataset.opt_file_1}')
log.info(f'SAR Image Year 0:{dataset.sar_file_0}')
log.info(f'SAR Image Year 1:{dataset.sar_file_1}')
#print(f'CMAP Image Year 0:{dataset.cmap_file_0}')
#print(f'CMAP Image Year 1:{dataset.cmap_file_1}')
log.info(f'Prev Def Image Year 1:{dataset.prev_def_file}')
pred_global_sum = np.zeros(dataset.original_shape+(general.N_CLASSES,))
t0 = time.perf_counter()
for overlap in tqdm(overlaps, leave=False, desc='Overlap'):
log.info(f'Predicting overlap {overlap}')
dataset.gen_patches(overlap = overlap)
dataloader = DataLoader(dataset, batch_size=args.batch_size, shuffle=False)
pbar = tqdm(dataloader, desc='Prediction', leave = False)
#preds = None
preds = torch.zeros((len(dataset), general.N_CLASSES, general.PATCH_SIZE, general.PATCH_SIZE))
for i, X in enumerate(pbar):
with torch.no_grad():
preds[args.batch_size*i: args.batch_size*(i+1)] = model(X).to('cpu')
preds = np.moveaxis(preds.numpy().astype(np.float16), 1, -1)
pred_sum = np.zeros(dataset.padded_shape+(general.N_CLASSES,)).reshape((-1, general.N_CLASSES))
pred_count = np.zeros(dataset.padded_shape+(general.N_CLASSES,)).reshape((-1, general.N_CLASSES))
for idx, idx_patch in enumerate(tqdm(dataset.idx_patches, desc = 'Rebuild', leave = False)):
crop_val = general.PREDICTION_REMOVE_BORDER
idx_patch_crop = idx_patch[crop_val:-crop_val, crop_val:-crop_val]
pred_sum[idx_patch_crop] += preds[idx][crop_val:-crop_val, crop_val:-crop_val]
pred_count[idx_patch_crop] += one_window[crop_val:-crop_val, crop_val:-crop_val]
#pred_sum[idx_patch] += preds[idx]
#pred_count[idx_patch] += one_window
pred_sum = pred_sum.reshape(dataset.padded_shape+(general.N_CLASSES,))
pred_count = pred_count.reshape(dataset.padded_shape+(general.N_CLASSES,))
pred_sum = pred_sum[general.PATCH_SIZE:-general.PATCH_SIZE,general.PATCH_SIZE:-general.PATCH_SIZE,:]
pred_count = pred_count[general.PATCH_SIZE:-general.PATCH_SIZE,general.PATCH_SIZE:-general.PATCH_SIZE,:]
pred_global_sum += pred_sum / pred_count
p_time = (time.perf_counter() - t0)/60
total_time += p_time
n_images += 1
log.info(f'Prediction time: {p_time} mins')
pred_global = pred_global_sum / len(overlaps)
#pred_b = pred_global.argmax(axis=-1).astype(np.uint8)
#pred_b[label == 2] = 2
#np.save(os.path.join(predicted_path, f'{general.PREDICTION_PREFIX}_{img_pair[0]}_{img_pair[1]}_{model_idx}.npy'), pred_b)
np.save(os.path.join(predicted_path, f'{general.PREDICTION_PREFIX}_prob_{img_pair[0]}_{img_pair[1]}_{model_idx}.npy'), pred_global[:,:,1].astype(np.float16))
#save_geotiff(str(args.base_image), os.path.join(visual_path, f'{general.PREDICTION_PREFIX}_{args.experiment}_{img_pair[0]}_{img_pair[1]}_{model_idx}.tif'), pred_b, dtype = 'byte')
pred_b2 = (pred_global[:,:,1] > 0.5).astype(np.uint8)
pred_b2[label == 2] = 2
base_image = os.listdir(str(args.base_image_path))[0]
base_data = os.path.join(str(args.base_image_path), base_image)
save_geotiff(base_data, os.path.join(visual_path, f'{general.PREDICTION_PREFIX}_{args.experiment}_{img_pair[0]}_{img_pair[1]}_{model_idx}.tif'), pred_b2, dtype = 'byte')
#save_geotiff(str(args.base_image), os.path.join(visual_path, f'{general.PREDICTION_PREFIX}_probs_{args.experiment}_{img_pair[0]}_{img_pair[1]}_{model_idx}.tif'), pred_global, dtype = 'float')
m_time = total_time / n_images
log.info(f'Mean Prediction time: {m_time} mins')
if __name__=="__main__":
for model_idx in range(args.number_models):
print(f'Predicting model {model_idx}')
p = Process(target=run, args=(model_idx,))
p.start()
p.join()