forked from lucasjinreal/keras_frcnn
-
Notifications
You must be signed in to change notification settings - Fork 1
/
detector.py
150 lines (126 loc) · 6.17 KB
/
detector.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
from __future__ import division
import os
import cv2
import numpy as np
import pickle
import time
from keras_frcnn import config
from keras_frcnn.config import Config
from keras import backend as K
from keras.layers import Input
from keras.models import Model
from keras_frcnn import roi_helpers
import argparse
import os
import keras_frcnn.resnet as nn
from keras_frcnn.visualize import draw_boxes_and_label_on_image_cv2
from utils.process import *
class FasterRCNNDetector(object):
def __init__(self, model_path):
self.model_path = model_path
if os.path.exists('config.pickle'):
with open('config.pickle', 'rb') as f:
self.cfg = pickle.load(f)
else:
self.cfg = Config()
print('Not found previous train and saved config.pickle file. may lose class map info.')
self._init_model()
def _init_model(self):
self.cfg.use_horizontal_flips = False
self.cfg.use_vertical_flips = False
self.cfg.rot_90 = False
class_mapping = self.cfg.class_mapping
if 'bg' not in class_mapping:
class_mapping['bg'] = len(class_mapping)
self.class_mapping = {v: k for k, v in class_mapping.items()}
input_shape_img = (None, None, 3)
input_shape_features = (None, None, 1024)
img_input = Input(shape=input_shape_img)
roi_input = Input(shape=(self.cfg.num_rois, 4))
feature_map_input = Input(shape=input_shape_features)
shared_layers = nn.nn_base(img_input, trainable=False)
# Define the RPN, built on the base layers
num_anchors = len(self.cfg.anchor_box_scales) * len(self.cfg.anchor_box_ratios)
rpn_layers = nn.rpn(shared_layers, num_anchors)
classifier = nn.classifier(feature_map_input, roi_input, self.cfg.num_rois,
nb_classes=len(class_mapping), trainable=True)
self.model_rpn = Model(img_input, rpn_layers)
model_classifier_only = Model([feature_map_input, roi_input], classifier)
self.model_classifier = Model([feature_map_input, roi_input], classifier)
model_path = self.cfg.model_path
print('Loading weights from {}'.format(model_path))
if not os.path.exists(model_path):
model_path = self.model_path
print('previous model path not found or not exist, using specific one: ', self.model_path)
self.model_rpn.load_weights(model_path, by_name=True)
self.model_classifier.load_weights(model_path, by_name=True)
self.model_rpn.compile(optimizer='sgd', loss='mse')
self.model_classifier.compile(optimizer='sgd', loss='mse')
def detect_on_image(self, img):
tic = time.time()
X, ratio = format_img(img, self.cfg)
# -if K.image_dim_ordering() == 'tf':
if K.image_data_format() == 'channels_last':
X = np.transpose(X, (0, 2, 3, 1))
# get the feature maps and output from the RPN
[Y1, Y2, F] = self.model_rpn.predict(X)
# this is result contains all boxes, which is [x1, y1, x2, y2]
# -result = roi_helpers.rpn_to_roi(Y1, Y2, self.cfg, K.image_dim_ordering(), overlap_thresh=0.7)
result = roi_helpers.rpn_to_roi(Y1, Y2, self.cfg, K.image_data_format(), overlap_thresh=0.7)
# convert from (x1,y1,x2,y2) to (x,y,w,h)
result[:, 2] -= result[:, 0]
result[:, 3] -= result[:, 1]
bbox_threshold = 0.8
# Apply the spatial pyramid pooling to the proposed regions
boxes = dict()
for jk in range(result.shape[0] // self.cfg.num_rois + 1):
rois = np.expand_dims(result[self.cfg.num_rois * jk:self.cfg.num_rois * (jk + 1), :], axis=0)
if rois.shape[1] == 0:
break
if jk == result.shape[0] // self.cfg.num_rois:
# pad R
curr_shape = rois.shape
target_shape = (curr_shape[0], self.cfg.num_rois, curr_shape[2])
rois_padded = np.zeros(target_shape).astype(rois.dtype)
rois_padded[:, :curr_shape[1], :] = rois
rois_padded[0, curr_shape[1]:, :] = rois[0, 0, :]
rois = rois_padded
[p_cls, p_regr] = self.model_classifier_only.predict([F, rois])
for ii in range(p_cls.shape[1]):
if np.max(p_cls[0, ii, :]) < bbox_threshold or np.argmax(p_cls[0, ii, :]) == (p_cls.shape[2] - 1):
continue
cls_num = np.argmax(p_cls[0, ii, :])
if cls_num not in boxes.keys():
boxes[cls_num] = []
(x, y, w, h) = rois[0, ii, :]
try:
(tx, ty, tw, th) = p_regr[0, ii, 4 * cls_num:4 * (cls_num + 1)]
tx /= self.cfg.classifier_regr_std[0]
ty /= self.cfg.classifier_regr_std[1]
tw /= self.cfg.classifier_regr_std[2]
th /= self.cfg.classifier_regr_std[3]
x, y, w, h = roi_helpers.apply_regr(x, y, w, h, tx, ty, tw, th)
except Exception as e:
print(e)
pass
boxes[cls_num].append(
[self.cfg.rpn_stride * x, self.cfg.rpn_stride * y, self.cfg.rpn_stride * (x + w), self.cfg.rpn_stride * (y + h),
np.max(p_cls[0, ii, :])])
# Add some nms to reduce many boxes
for cls_num, box in boxes.items():
boxes_nms = roi_helpers.non_max_suppression_fast(box, overlap_thresh=0.5)
boxes[cls_num] = boxes_nms
print(self.class_mapping[cls_num] + ":")
for b in boxes_nms:
b[0], b[1], b[2], b[3] = get_real_coordinates(ratio, b[0], b[1], b[2], b[3])
print('{} prob: {}'.format(b[0: 4], b[-1]))
img = draw_boxes_and_label_on_image_cv2(img, self.class_mapping, boxes)
print('Elapsed time = {}'.format(time.time() - tic))
cv2.imshow('image', img)
# Change results_images to resust_images
result_path = './result_images/{}.png'.format('result')
print('result saved into ', result_path)
cv2.imwrite(result_path, img)
cv2.waitKey(0)
def detect_on_video(self, v):
pass