-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtfckpt2keras_inference.py
211 lines (165 loc) · 7.31 KB
/
tfckpt2keras_inference.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
from tensorflow.keras.preprocessing import image
from tensorflow.keras.optimizers import Adam
from imageio import imread
import numpy as np
from matplotlib import pyplot as plt
from losses.keras_ssd_loss import SSDLoss
from models.mbv3ssdlite_model import MobileNetV3SSDLite
import os
import tensorflow as tf
import h5py
import tensorflow.keras.backend as K
from PIL import Image
from object_detection.utils import ops as utils_ops
from object_detection.utils import label_map_util
# os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
# os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
# model config
batch_size = 32
image_size = (320, 320, 3)
n_classes = 90
mode = 'inference'
l2_regularization = 0.0004
min_scale = 0.2 #None
max_scale = 0.95 #None
scales = [0.1, 0.35, 0.5, 0.65, 0.8, 0.95, 1.0]
aspect_ratios_global = None #[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0] # None
aspect_ratios_per_layer = [[1.01, 2.0, 0.5], [1.0, 2.0, 0.5, 3.0, 1.0 / 3.0], [1.0, 2.0, 0.5, 3.0, 1.0 / 3.0],
[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0], [1.0, 2.0, 0.5, 3.0, 1.0 / 3.0], [1.0, 2.0, 0.5, 3.0, 1.0 / 3.0]]
two_boxes_for_ar1 = True
steps = None #[16, 32, 64, 107, 160, 320]
offsets = None #[0.5, 0.5, 0.5, 0.5, 0.5, 0.5]
clip_boxes = False #True
variances = [1.0,1.0,1.0,1.0]#[0.1, 0.1, 0.1, 0.1]#[0.1, 0.1, 0.2, 0.2]
coords = 'centroids'
normalize_coords = True
subtract_mean = [0,0,0] #[123, 117, 104]
divide_by_stddev = 1 #28
swap_channels = None
confidence_thresh = 0.5 # 0.65
iou_threshold = 0.6
top_k = 100 #200
nms_max_output_size = 100 #400
return_predictor_sizes = False
model_type="small_extractor"
width_multiplier=1.0
model_name="FeatureExtractor"#"Mobilenetv3_Large"
divisible_by=8
K.clear_session()
K.set_learning_phase(0)
model = MobileNetV3SSDLite(image_size, n_classes, mode, l2_regularization, min_scale, max_scale, scales,
aspect_ratios_global, aspect_ratios_per_layer, two_boxes_for_ar1, steps,
offsets, clip_boxes, variances, coords, normalize_coords, subtract_mean,
divide_by_stddev, swap_channels, confidence_thresh, iou_threshold, top_k,
nms_max_output_size, return_predictor_sizes,model_type,width_multiplier,model_name,divisible_by)
input_tensor = tf.keras.layers.Input(shape=image_size)
output_tensor = model(input_tensor)
# ,input_shape=image_size
model = tf.keras.Model(inputs=[model.input],outputs=[model.output],trainable=False)
print(model.summary())
# FeatureExtractor
print(model.layers[1].summary())
print(model.layers[1].layers[3].summary())
# # MobileNetv3
print(model.layers[1].layers[3].layers[0].summary())
# 2: Load the trained weights into the model.
initial_weights = [layer.get_weights() for layer in model.layers]
# weights_path = './ckpt/tf_convert_mbv3ssdlite_large_0528.h5'
# weights_path = './ckpt/tf_convert_mbv3ssdlite_small_0527.h5'
def dfsmodel(prefix,layers,f):
# for key,value in f.items():
# print("cur f:",key,value)
for layer in layers:
# print(prefix + '/' +layer.name)
if hasattr(layer, 'layers'):
layer = dfsmodel(prefix+'/'+layer.name,layer.layers,f[layer.name])
else:
print(prefix + '/' +layer.name)
if(len(layer.weights)>0):
print(layer.name,f[layer.name])
layer.set_weights(f[layer.name])
# global weights_num
# weights_num+=1
print("loaded {} weights".format(prefix + '/' +layer.name))
return layers
# load weights
# h5fn = h5py.File(weights_path,'r')
# dfsmodel('',model.layers[1].layers,h5fn)
# model.load_weights("./ckpt/loaded_large_0528_weights.h5",by_name=True)
model.load_weights("./ckpt/loaded_small_0527_weights.h5",by_name=True)
for layer, initial in zip(model.layers, initial_weights):
weights = layer.get_weights()
print(layer.name)
print("len(weights)",len(weights))
# print(initial)
if weights and all(tf.nest.map_structure(np.array_equal, weights, initial)):
print(f'first loaded contained no weights for layer {layer.name}!')
# h5 weights loaded successfully
# save weights
# model.save_weights("./ckpt/loaded_large_0528_weights.h5")
# model.save_weights("./ckpt/loaded_small_0527_weights.h5")
# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = './testfile/mscoco_label_map.pbtxt'
category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True)
def load_image_into_numpy_array(image):
(im_width, im_height) = image.size
imgdata = np.array(image.getdata())
if(len(imgdata.shape)==1):
return imgdata.reshape((im_height, im_width, 1)).astype(np.uint8).repeat(3,axis=2)
return imgdata.reshape(
(im_height, im_width, 3)).astype(np.uint8)
# We'll only load one image in this example.
img_dir = './imgs/test'
for imgfn in os.listdir(img_dir):
img_path = os.path.join(img_dir, imgfn)
print(img_path)
orig_images = [] # Store the images here.
input_images = [] # Store resized versions of the images here.
orig_images.append(imread(img_path))
# keras preprocess
# img = image.load_img(img_path, target_size=(image_size[0], image_size[1]))
# img = image.img_to_array(img)
# input_images.append(img)
image = Image.open(img_path)
image = image.resize((image_size[0],image_size[1]),Image.ANTIALIAS)
image_np = load_image_into_numpy_array(image)
input_images.append(image_np)
input_images = np.array(input_images,dtype=np.float32)
print("origin image",input_images)
# tf preprocess
input_images = input_images*(2.0 / 255.0)-1.0
print("image",input_images)
# outputs = model.predict(input_images)
y_pred = model.predict(input_images)
print("predict end.")
#confidence_threshold = 0.5
confidence_threshold = confidence_thresh
y_pred_thresh = [y_pred[k][y_pred[k,:,1] > confidence_threshold] for k in range(y_pred.shape[0])]
np.set_printoptions(precision=2, suppress=True, linewidth=90)
print("Predicted boxes:\n")
print(' class conf xmin ymin xmax ymax')
print("y_pred_thresh[0]",y_pred_thresh[0])
colors = plt.cm.hsv(np.linspace(0, 1, 91)).tolist()
plt.figure(figsize=(20, 12))
plt.imshow(orig_images[0])
current_axis = plt.gca()
# print("y_pred_thresh[0]",y_pred_thresh[0])
for box in y_pred_thresh[0]:
# Transform the predicted bounding boxes for the 300x300 image to the original image dimensions.
xmin = box[2] * orig_images[0].shape[1] / image_size[0]
ymin = box[3] * orig_images[0].shape[0] / image_size[1]
xmax = box[4] * orig_images[0].shape[1] / image_size[1]
ymax = box[5] * orig_images[0].shape[0] / image_size[0]
# print("xmin:{},ymin:{},xmax:{},ymax:{}".format(xmin,ymin,xmax,ymax))
color = colors[int(box[0])]
if(int(box[0]) not in category_index.keys()):
continue
clsname = category_index[int(box[0])]['name']
label = '{}: {:.2f}'.format(clsname, box[1])
current_axis.add_patch(plt.Rectangle((xmin, ymin), xmax-xmin, ymax-ymin, color=color, fill=False, linewidth=2))
current_axis.text(xmin, ymin, label, size='x-large', color='white', bbox={'facecolor':color, 'alpha':1.0})
#plt.show()
plt.savefig("./imgs/detections_test/keras_{}".format(imgfn))
plt.ion()
plt.pause(2)
plt.close()