-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathIP_test.py
248 lines (219 loc) · 13.1 KB
/
IP_test.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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
# Copyright (c) 2021-2022 Alibaba Group Holding Limited.
import os
import sys
import glob
import argparse
import numpy as np
from time import time
from PIL import Image
import tensorflow as tf
from datetime import datetime
from IP_config import PConfig
from networks.IP_model import PModel
from utils.dataloader import Dataloader
from utils.utils import load_single_image, str2bool
def parse_args():
parser = argparse.ArgumentParser(description='Train a detector')
parser.add_argument('--loss_metric', type=str, default="PSNR", help='loss_metric: PSNR or SSIM')
parser.add_argument('--model_name', type=str, default="STPM", help='loss_metric: PSNR or SSIM')
parser.add_argument('--work_dir', type=str, default=None, help='the dir to save logs and models, load the models')
parser.add_argument('--is_post', type=str2bool, default=True, help='add the Unet post network')
parser.add_argument('--with_context_model', type=str2bool, default=True, help='add the context model network')
parser.add_argument('--is_multi', type=str2bool, default=True, help='enable variable rate control')
parser.add_argument('--seed', type=int, default=1000, help='random seed')
parser.add_argument('--idx_test', type=int, default=0, help='idx_test')
parser.add_argument('--ckpt_dir_pre', type=str, default=None, help='I-frame pretrained models are needed for P-frame compression')
args = parser.parse_args()
return args
def main(unused_argv):
args = parse_args()
PConfig.cckpt(args)
print(args)
print(PConfig)
#model and test config
os.environ['CUDA_VISIBLE_DEVICES']=','.join(["%d"%id for id in PConfig.gpus_test])
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
batch_size = 1
PConfig.fix_size =[1, 1080, 1920*3, 3]
# PConfig.test_set_dir = "../data/1080P/"
# PConfig.test_comment = "test_for_valid"
#built the model
PMod=PModel(is_train=False)
graph, sess = PMod.build()
if not os.path.isdir(PConfig.info_dir):
os.makedirs(PConfig.info_dir)
alpha_list = [1, 0.9, 0.7, 0.5, 0.3, 0.1]
#open graph context manager
print("imgaeGenerator")
with graph.as_default():
bpp_avg = 0 # 不同序列间测试均值
bpp_res_avg = 0
bpp_min_avg = 0
psnr_avg = 0
ssim_avg = 0
PConfig.test_seq_names.sort()
for seq_name in PConfig.test_seq_names:
bpp_rd = [] # 多码率用来保存不同lambda下的结果
psnr_rd = []
ssim_rd = []
bpp_res_rd = []
bpp_min_rd = []
print("the test seq_name is %s" % seq_name)
file = open(PConfig.datatext_dir, 'a+')
file.write("\n#######################################################\n")
file.write("Below is %s sequence test results (1I%dP)\n"%(seq_name, PConfig.test_GOP-1))
file.write("#######################################################\n")
file.close()
ori_seq_dir = os.path.join(PConfig.test_set_dir, seq_name)
# I_seq_dir = os.path.join(PConfig.test_I_dir, seq_name)
# I_pathes = glob.glob(os.path.join(I_seq_dir, "*.png"))
# I_pathes.sort()
# info_seq_path = os.path.join(PConfig.test_I_dir, "info", "%s.npy"%seq_name)
# info_seq = np.load(info_seq_path)
test_loader = Dataloader(ori_seq_dir, 8, batch_size, model_type="test") # pipeline的方式
test_len = test_loader.file_len
test_filenames = test_loader.test_filenames
init_test = test_loader.initializer
test_next = test_loader.test_image_batch
for index_random in range(len(PConfig.lambda_list)-1):
for alpha_rand in alpha_list:
# alpha_rand = 0.7
if PConfig.is_multi:
# l_onehot = alpha_rand*PConfig.lambda_onehot[index_random] + (1-alpha_rand)*PConfig.lambda_onehot[index_random+1]
# lambda_test = alpha_rand*PConfig.lambda_list[index_random] + (1-alpha_rand)*PConfig.lambda_list[index_random+1]
l_onehot = PConfig.lambda_onehot[PConfig.idx_test]
lambda_test = PConfig.lambda_list[PConfig.idx_test]
PConfig.test_comment = "lambda_%d"%(lambda_test)
else:
lambda_test = PConfig.train_lambda
PConfig.test_comment = "lambda_%d"%(lambda_test)
sess.run(init_test) # 每次都重新开始初始化test iteration迭代器，重头开始
average_psnr = 0.
average_bpp = 0.
average_estbpp = 0.
average_estbpp_res = 0.
average_estbpp_min = 0.
average_msssim = 0.
if not os.path.isdir(PConfig.rescon_dir):
os.makedirs(PConfig.rescon_dir)
if not os.path.isdir(PConfig.bin_dir):
os.makedirs(PConfig.bin_dir)
if not os.path.isdir(os.path.join(PConfig.rescon_dir, seq_name)):
os.makedirs(os.path.join(PConfig.rescon_dir, seq_name))
#######################################################################
for i in range(test_len):
# for i in range(3):
tic = time()
if i%PConfig.test_GOP == 0: # I帧直接读取已经完成的编解码信息
frame_flag = "I"
cur_img_batch = sess.run(test_next)
image_name = os.path.basename(test_filenames[i])
input_image_batch = np.concatenate((cur_img_batch, cur_img_batch, cur_img_batch), axis=2)
if PConfig.is_multi:
feed_dict={PMod.input_image_in:input_image_batch, PMod.lambda_onehot:l_onehot}
else:
feed_dict={PMod.input_image_in:input_image_batch}
psnr, ms_ssim, bpp, bpp_y = sess.run([PMod.psnr, PMod.ms_ssim, PMod.bpp, PMod.bpp_y], feed_dict=feed_dict)
bpp_res = bpp
bpp_min = np.minimum(bpp, bpp_res)
else:
frame_flag = "P"
cur_img_batch = sess.run(test_next)
# pre_img_recon_batch = clipped_recon_image
image_shape = np.shape(cur_img_batch)
image_name = os.path.basename(test_filenames[i])
image_recon_path = os.path.join(PConfig.rescon_dir, seq_name, image_name)
input_image_batch = np.concatenate((pre_img_batch, pre_img_batch, cur_img_batch), axis=2)
if PConfig.is_multi:
feed_dict={PMod.input_image_in:input_image_batch, PMod.lambda_onehot:l_onehot}
else:
feed_dict={PMod.input_image_in:input_image_batch}
psnr, ms_ssim, bpp, bpp_res, bpp_y_res, recon_image = sess.run([PMod.psnr, PMod.ms_ssim, PMod.bpp, PMod.bpp_res, PMod.bpp_y_res, PMod.clip_recon_image], feed_dict=feed_dict)
tic1 = time()
bpp_min = np.minimum(bpp, bpp_res)
print(bpp_res, bpp_y_res, bpp_res-bpp_y_res)
#########################调用最新的编码文件###################################################
# bin_path = PConfig.bin_dir + image_name.replace(".png",".bin")
actual_total_bits = 0 # 这里还没有开始写实际编解码函数-entropy_encoding
actual_bpp = actual_total_bits / (batch_size * image_shape[1] * image_shape[2])
tic2 = time()
# 将训练后的图像保存到data-recon-mini512中
# print('The shape is Recon', clipped_recon_image.shape)
clipped_recon_image = (np.round(recon_image*255)).astype(np.uint8)
Image.fromarray(clipped_recon_image[0]).save(image_recon_path)
# nowtime = datetime.now().strftime('%Y-%m-%d %H:%M:%2S')
# print("%s %s : the nn time is %.2f s, the entropy time is %.2f s "%(nowtime, image_name, tic1-tic, tic2-tic1), image_shape)
pre_img_batch = cur_img_batch
average_psnr += psnr
# average_bpp += actual_bpp
average_estbpp += bpp
average_estbpp_res += bpp_res
average_estbpp_min += bpp_min
average_msssim += ms_ssim
print("%dth frame (%s) result is %.4f, %.4f, %.4f, %.4f, %.4f (bpp, bpp_res, bpp_min, psnr, msssim)"%(i, frame_flag, bpp, bpp_res, bpp_min, psnr, ms_ssim))
# txt_write[image_name] = [psnr_val, ms_ssim_np, 0, bpp_val]
if i < 10:
file = open(PConfig.datatext_dir, 'a+')
file.write("%dth frame (%s) result is %.4f, %.4f, %.4f, %.4f, %.4f (bpp, bpp_res, bpp_min, psnr, msssim)\n"%(i, frame_flag, bpp, bpp_res, bpp_min, psnr, ms_ssim))
file.close()
bpp_rd.append(average_estbpp / test_len)
bpp_res_rd.append(average_estbpp_res / test_len)
bpp_min_rd.append(average_estbpp_min / test_len)
psnr_rd.append(average_psnr / test_len)
ssim_rd.append(average_msssim / test_len)
print("lambda:%d, %s sequence test result, PSNR:%.3f, msssim:%.4f, Ibpp:%.4f, Resbpp:%.4f, Minbpp:%.4f\n\n" % (
lambda_test, seq_name,
average_psnr / test_len,
average_msssim / test_len,
average_estbpp / test_len,
average_estbpp_res / test_len,
average_estbpp_min / test_len))
file = open(PConfig.datatext_dir, 'a+')
file.write("the number of PMod is %s %s\n" % (PMod.module_file, PConfig.test_comment))
file.write("%s sequence test result is PSNR:%.3f, msssim:%.4f, Ibpp:%.4f, Resbpp:%.4f, Minbpp:%.4f\n\n" % (seq_name,
average_psnr / test_len,
average_msssim / test_len,
average_estbpp / test_len,
average_estbpp_res / test_len,
average_estbpp_min / test_len))
file.close()
## 在非多码率模型时，要跳出多码率的循环
# if not PConfig.is_multi:
break
# if not PConfig.is_multi:
break
# if PConfig.is_multi:
# file1 = open(PConfig.info_dir+"RD_%s.txt"%seq_name, 'a+')
# file1.write("the number of PMod is %s\nthe seq name is %s:\nbpp = [" % (PMod.module_file, seq_name))
# for bpp_s in bpp_rd:
# file1.write("%.4f, "%(bpp_s))
# file1.write("];\nPSNR = [")
# for psnr_s in psnr_rd:
# file1.write("%.3f, "%(psnr_s))
# file1.write("];\nSSIM = [")
# for ssim_s in ssim_rd:
# file1.write("%.4f, "%(ssim_s))
# file1.write("];\n\n")
# file1.close()
bpp_avg += bpp_rd[0]
bpp_res_avg += bpp_res_rd[0]
bpp_min_avg += bpp_min_rd[0]
psnr_avg += psnr_rd[0]
ssim_avg += ssim_rd[0]
print("lambda:%d, the final test result is %.4f, %.4f, %.4f, %.4f, %.4f (bpp, bpp_res, bpp_min, psnr, msssim)\n\n" % (
lambda_test,
bpp_avg / len(PConfig.test_seq_names),
bpp_res_avg / len(PConfig.test_seq_names),
bpp_min_avg / len(PConfig.test_seq_names),
psnr_avg / len(PConfig.test_seq_names),
ssim_avg / len(PConfig.test_seq_names)))
file = open(PConfig.datatext_dir, 'a+')
file.write("the final test result is %.4f, %.4f, %.4f, %.4f, %.4f (bpp, bpp_res, bpp_min, psnr, msssim)\n\n" % (
bpp_avg / len(PConfig.test_seq_names),
bpp_res_avg / len(PConfig.test_seq_names),
bpp_min_avg / len(PConfig.test_seq_names),
psnr_avg / len(PConfig.test_seq_names),
ssim_avg / len(PConfig.test_seq_names)))
file.close()
if __name__ == '__main__':
tf.app.run()