forked from eriklindernoren/PyTorch-YOLOv3
-
Notifications
You must be signed in to change notification settings - Fork 0
/
detect.py
216 lines (150 loc) · 6.91 KB
/
detect.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
from __future__ import division
from models import *
from utils.utils import *
from utils.datasets import *
import click
from tqdm import tqdm
import torch
from torch.utils.data import DataLoader
import pandas as pd
import sqlite3
import io
from PIL import Image
class SqliteDataset(Dataset):
def __init__(self, sqlite_file, img_size, table_name=None):
super(Dataset, self).__init__()
self.img_size = img_size
if table_name is None:
self.table_name = "images"
else:
self.table_name = table_name
self.conn = None
self.sqlite_file = sqlite_file
with sqlite3.connect(sqlite_file) as conn:
self.length = conn.execute("SELECT count(*) FROM {}".format(self.table_name)).fetchone()[0]
print("SQLITE Dataset of size {}.".format(self.length))
def __getitem__(self, index):
if self.conn is None:
self.conn = sqlite3.connect(self.sqlite_file)
self.conn.execute('pragma journal_mode=wal')
result = self.conn.execute("SELECT filename, data, scale_factor FROM {} WHERE rowid=?".format(self.table_name),
(index,)).fetchone()
if result is not None:
filename, data, scale_factor = result
buffer = io.BytesIO(data)
img = Image.open(buffer).convert('RGB')
else:
filename = "ERROR-dummy.jpg"
scale_factor = 1.0
print('Something went wrong on image {}.'.format(index))
print('Providing dummy result ...')
img = Image.new('RGB', (256, 256))
img = transforms.ToTensor()(img)
img, _ = self.pad_to_square(img, 0)
img_size = np.array(img.shape)
img_size[1:] = img_size[1:] / scale_factor
img = resize(img, self.img_size)
return filename, img, img_size
def __len__(self):
return self.length
@staticmethod
def pad_to_square(img, pad_value):
c, h, w = img.shape
dim_diff = np.abs(h - w)
# (upper / left) padding and (lower / right) padding
pad1, pad2 = dim_diff // 2, dim_diff - dim_diff // 2
# Determine padding
pad = (0, 0, pad1, pad2) if h <= w else (pad1, pad2, 0, 0)
# Add padding
img = F.pad(img, pad, "constant", value=pad_value)
return img, pad
@click.command()
@click.argument('image_folder', type=click.Path(exists=True))
@click.argument('model_def', type=click.Path(exists=True))
@click.argument('weights_path', type=click.Path(exists=True))
@click.argument('result_file', type=click.Path())
# @click.argument('class_path', type=click.Path(exists=True))
@click.option('--batch-size', type=int, default=16, help="size of the batches")
@click.option("--img-size", type=int, default=416, help="size of each image dimension")
@click.option('--n-cpu', type=int, default=4, help="number of cpu threads to use during batch generation")
@click.option("--conf-thres", type=float, default=0.01, help="object confidence threshold")
@click.option("--sqlite-table-name", type=str, default=None, help="")
# @click.option("--nms-thres", type=float, default=0.4, help="iou threshold for non-maximum suppression")
# @click.option('--perform-nms', type=bool, is_flag=True, default=False,
# help="add if you actually want non-maximum supression")
def cli(image_folder, model_def, weights_path, result_file, batch_size, img_size, n_cpu, conf_thres, sqlite_table_name):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Set up model
model = Darknet(model_def, img_size=img_size).to(device)
if weights_path.endswith(".weights"):
# Load darknet weights
model.load_darknet_weights(weights_path)
else:
# Load checkpoint weights
model.load_state_dict(torch.load(weights_path))
model.eval() # Set in evaluation mode
if os.path.isdir(image_folder):
dataloader = DataLoader(
ImageFolder(image_folder, img_size=img_size),
batch_size=batch_size,
shuffle=False,
num_workers=n_cpu,
)
else:
dataloader = DataLoader(
SqliteDataset(image_folder, img_size=img_size, table_name=sqlite_table_name),
batch_size=batch_size,
shuffle=False,
num_workers=n_cpu,
)
# noinspection PyUnresolvedReferences
Tensor = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
results = []
print("\nPerforming object detection:")
progress = tqdm(enumerate(dataloader), total=len(dataloader))
num_found = 0
check_point = 10000
for batch_i, (img_paths, input_imgs, img_sizes) in progress:
# Configure input
input_ten = Variable(input_imgs.type(Tensor))
img_sizes = img_sizes.cpu().numpy()
# Get detections
with torch.no_grad():
output = model(input_ten)
tmp = output.cpu().numpy()
detections = []
for i, part in enumerate(np.split(tmp, len(tmp))):
detections.append(pd.DataFrame(part[..., :5].squeeze(),
columns=['center_x', 'center_y', 'box_w', 'box_h', 'conf'],
index=len(part.squeeze()) * [i]))
detections = pd.concat(detections)
detections['x1'] = detections.center_x - detections.box_w / 2.0
detections['y1'] = detections.center_y - detections.box_h / 2.0
detections['x2'] = detections.center_x + detections.box_w / 2.0
detections['y2'] = detections.center_y + detections.box_h / 2.0
detections = detections.reset_index().rename(columns={'index': 'image'})
detections['im_width'] = img_sizes[detections.image, 1]
detections['im_height'] = img_sizes[detections.image, 2]
detections = detections.loc[detections.conf > conf_thres]
if len(detections) == 0:
continue
detections =\
pd.concat(
[pd.DataFrame(
rescale_boxes(dpart[['x1', 'y1', 'x2', 'y2', 'conf']].values,
img_size, [dpart.im_width.unique(), dpart.im_height.unique()]),
index=len(dpart) * [img_paths[i]],
columns=['x1', 'y1', 'x2', 'y2', 'conf']) for i, dpart in detections.groupby('image')]).\
reset_index().rename(columns={'index': 'path'})
detections['box_w'] = detections.x2 - detections.x1
detections['box_h'] = detections.y2 - detections.y1
num_found += len(detections)
progress.set_description("Num found: {}".format(num_found))
results.append(detections[['path', 'x1', 'y1', 'box_w', 'box_h', 'conf']])
del input_imgs
del input_ten
del output
if num_found > check_point:
pd.concat(results).to_pickle(result_file)
check_point = num_found + 10000
pd.concat(results).to_pickle(result_file)