-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluate.py
128 lines (103 loc) · 4.51 KB
/
evaluate.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
import os
import argparse
import warnings
import json
import numpy as np
import torch.cuda
from modules.utils import get_subset_sampler, plot_img, get_reverse_imagenet_transform
from modules.model import DeepGazeVGG
from modules.dataset import SaliconDataset
from torch.utils.data import DataLoader, RandomSampler
from PIL import Image
from pathlib import Path
from tqdm import tqdm
def load_model_from_file(weight_file, config):
model = DeepGazeVGG(**config['model'])
state_dict = torch.load(weight_file)
model.load_state_dict(state_dict)
model = model.eval()
if torch.cuda.is_available():
model = model.cuda()
return model
def evaluate_model(args, config, should_plot=True):
model = load_model_from_file(args.weight_file, config)
test_dataset = SaliconDataset(args.data_path, ['valid'])
if config['max_images'] is not None:
test_sampler = get_subset_sampler(test_dataset, config['max_images'])
else:
test_sampler = RandomSampler(test_dataset)
train_loader = DataLoader(test_dataset, batch_size=1, sampler=test_sampler, num_workers=0)
device = torch.device("cuda") if torch.cuda.is_available() else "cpu"
trans = get_reverse_imagenet_transform()
for img, heatmap in tqdm(train_loader):
print("img[1] = ", img[1][0])
img_tensor = img[0].to(device)
print("img_tensor.shape = ", img_tensor.shape)
with torch.no_grad():
output = model(img_tensor)
pred_heatmap = output.detach().cpu().numpy()[0, 0, :, :]
img_tensor = trans(img_tensor.detach().cpu())
img_np = img_tensor.numpy()[0].transpose([1, 2, 0])
cropped_img = crop(img_np, pred_heatmap, aspect_ratio=args.aspect_ratio)
if should_plot:
plot_img(img_np, title="original")
plot_img(img_np * 255, mask=pred_heatmap, title="heatmap")
plot_img(cropped_img*255, title="crop")
Image.fromarray((img_np*255).astype(np.uint8)).save(Path(args.output_path) / Path(img[1][0]).name)
Image.fromarray((pred_heatmap * 255).astype(np.uint8)).save(Path(args.output_path) / Path(heatmap[1][0]).name)
Image.fromarray((cropped_img * 255).astype(np.uint8)).save(Path(args.output_path) / ("crop_"+Path(img[1][0]).name))
def crop(img, heatmap, aspect_ratio=1):
h, w = img.shape[:2]
# ar is w/h
if h > w:
if aspect_ratio < 1:
# crop height>crop width
crop_w = w
crop_h = int(w / aspect_ratio)
else:
# crop height<=crop width
crop_h = h
crop_w = int(h * aspect_ratio)
else:
# h<=w
if aspect_ratio < 1:
# crop height>crop width
crop_h = h
crop_w = int(h * aspect_ratio)
else:
# crop height<=crop width
crop_w = w
crop_h = int(w / aspect_ratio)
max_row, max_col, max_score = -1, -1, -1
for row in range(0, img.shape[0] - crop_h + 1):
row_end = row + crop_h
for col in range(0, img.shape[1] - crop_w + 1):
col_end = col + crop_w
curr_score = heatmap[row:row_end, col:col_end].sum()
if curr_score > max_score:
max_row = row
max_col = col
max_score = curr_score
return img[max_row:max_row + crop_h, max_col:max_col + crop_w]
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--config_file', type=str, required=True, help='path to configuration json')
parser.add_argument('--weight_file', type=str, required=True, help='trained pth file')
parser.add_argument('--output_path', type=str, required=True, help='results path')
parser.add_argument('--data_path', type=str, required=True, help='SALICON images folder')
parser.add_argument('--max_images', type=str, required=False, help='maximum number of images to use', default=3)
parser.add_argument('--center_bias_path', type=str, required=False, help='optional center bias numpy file')
parser.add_argument('--aspect_ratio', type=str, required=False, help='aspect ratio for crop', default=0.5)
args = parser.parse_args()
if not os.path.exists(args.output_path):
os.makedirs(args.output_path)
else:
warnings.warn('output_path already exist!')
with open(args.config_file) as data_file:
config = json.load(data_file)
if args.max_images is not None:
config['max_images'] = args.max_images
evaluate_model(args, config, should_plot=False)
print("Done!")
if __name__ == '__main__':
main()