def create_dataset_testfile(dir_path, out_path):
subdir_list = os.listdir(dir_path)
sampler = TestSampler()
plt.ion()
plt.figure(1)
for name in subdir_list:
print('start process {}:'.format(name))
cnt = 0
fp = open(os.path.join(out_path, str(name) + '.pkl'), 'wb')
img_namelist = os.listdir(os.path.join(dir_path, name))
img_namelist = np.sort(img_namelist)
for img in img_namelist:
img_path = os.path.join(dir_path, name, img)
img_data = cv2.imread(img_path, cv2.IMREAD_COLOR)
img_data = cv2.resize(img_data, (640, 480))
#img_data=np.copy(img_data[:,:,::-1])
# print(img_data.shape)
# print(img_data)
sample_list, scaled = sampler.sample_and_filter(img_data)
plt.imshow(scaled[:,:,::-1])
plt.draw()
plt.show()
obj = {'origin': utils.encode_img(img_data), 'scaled': utils.encode_img(scaled)}
pickle.dump(obj, fp, -1)
for sample in sample_list:
pickle.dump(sample, fp, -1)
cnt = cnt + 1
print(cnt)
fp.close()
def sampling_process(self,data,scaled,shape):
"""
采用随机性策略抽取样本，缺少样本数量的确定性，但是可以很大程度加快处理速度，视情况采用
:param 输入的原始图像矩阵，要求格式为（H，W，C）通道最好为RGB
:param scaled: 缩放（4倍）后的图像矩阵
:return: 抽样得到的样本列表
"""
sample_list = []
#if self.cache0 is None or self.cache1 is None:
#return sample_list
dx=dy=20
self.cache.append((data,scaled,shape))
for i in range(shape[0]//dy):
for j in range(shape[1]//dx):
#lr = scaled[i * dx:(i + 1) * dx, j * dy:(j + 1) * dy, :]
if np.random.rand()<self.sample_rate:#*self.get_weight(lr):
LQs = []
GT = []
#self.logger.info('cache len is {}'.format(len(self.cache)))
for index in range(len(self.cache)):
GT.append(self.cache[index][0][i * dx * 4:(i + 1) * dx * 4, j * dy * 4:(j + 1) * dy * 4, :])
if index%2==0:
LQs.append(self.cache[index][1][i * dx:(i + 1) * dx, j * dy:(j + 1) * dy, :])
sample = {'LQs': [utils.encode_img(lr) for lr in LQs], 'GT': [utils.encode_img(hr) for hr in GT]
,'sample':True}
sample_list.append(sample)
self.cache.clear()
self.cache.append((data,scaled,shape))
#chean sample cache
return sample_list
def create_PSNR_testfile(self,file_path):
"""
这部分或许可以直接在服务端验证
:param file_path:
:return:
"""
f_path=Path(file_path)
if not f_path.parent.exists():
f_path.parent.mkdir(parents=True)
sampler=TestSampler()
fp=open(file_path,'wb')
with picamera.PiCamera() as camera:
rawFrame = PiRGBArray(camera, (640, 480))
camera.resolution = (640, 480)
camera.framerate = 20
shape=(640,480)
time.sleep(2)
self.start = time.time()
cnt=0
start=time.time()
for frame in camera.capture_continuous(rawFrame, 'rgb', use_video_port=True):
origin = np.array(frame.array, dtype=np.uint8)
sample_list, scaled = sampler.sample_and_filter(origin)
h_n = int(80 * np.ceil(origin.shape[0] / 80))
w_n = int(80 * np.ceil(origin.shape[1] / 80))
normed = np.zeros((h_n, w_n, 3), dtype=np.float32) / 255.
normed[0:origin.shape[0], 0:origin.shape[1], :] = origin
obj = {'origin': utils.encode_img(normed), 'scaled': utils.encode_img(scaled),'shape':shape}
pickle.dump(obj, fp, -1)
for sample in sample_list:
pickle.dump(sample,fp,-1)
cnt = cnt + 1
if cnt%10:
self.logger.info("sent {} frames in {} sec,fps:{}".format(cnt, time.time() - start,
cnt / (time.time() - start)))
if cnt>200:
break
self.logger.info(cnt)
self.count = self.count + 1
rawFrame.seek(0)
rawFrame.truncate()
if self.terminate:
self.logger.info('terminate')
break
fp.close()
def create_datafile(self,file_path):
"""
create the original captured frames,encoded in png
:param file_path: the path of the data pkl file
"""
f_path=Path(file_path)
if not f_path.parent.exists():
f_path.parent.mkdir(parents=True)
sampler=TestSampler(sample_len=7)
fp=open(file_path,'wb')
with picamera.PiCamera() as camera:
rawFrame = PiRGBArray(camera, (640, 480))
camera.resolution = (640, 480)
camera.framerate = 15
shape=(640,480)
time.sleep(2)
self.start = time.time()
cnt=0
start=time.time()
prev_cnt=0
len_sum=0
for frame in camera.capture_continuous(rawFrame, 'rgb', use_video_port=True):
frame = np.array(frame.array, dtype=np.uint8)
#frame = cv2.resize(frame, dsize=(0, 0), fx=0.25, fy=0.25, interpolation=cv2.INTER_AREA)
# frame=utils.imresize_np(frame,1/4,True)
pickle.dump(utils.encode_img(frame), fp, -1)
# sample_list, scaled = sampler.sample_and_filter(np.array(frame.array, dtype=np.uint8))
# #self.logger.info('len of sample list is {}'.format(len(sample_list)))
# len_sum=len_sum+len(sample_list)
# obj={'img':utils.encode_img(scaled),'shape':shape,'sample':False}
# pickle.dump(obj,fp,-1)
#
# for sample in sample_list:
# pickle.dump(sample,fp,-1)
# # self.logger.info('send sample')
cnt = cnt + 1
self.count = self.count + 1
if cnt%10==0:
self.logger.info('total sample num is:{}'.format(len_sum))
self.logger.info("sent {} frames in {} sec,fps:{}".format(cnt, time.time() - start,
cnt / (time.time() - start)))
start=time.time()
cnt=0
len_sum=0
if self.count>200:
break
#self.logger.info(cnt)
rawFrame.seek(0)
rawFrame.truncate()
if self.terminate:
self.logger.info('terminate')
break
fp.close()
def create_dataset_testfile(self,dir_path,out_path):
subdir_list=os.listdir(dir_path)
sampler=TestSampler()
plt.ion()
plt.figure(1)
for name in subdir_list:
self.logger.info('start process {}:'.format(name))
cnt=0
fp=open(os.path.join(out_path,str(name)+'.pkl'),'wb')
img_namelist=os.listdir(os.path.join(dir_path,name))
img_namelist= np.sort(img_namelist)
for img in img_namelist:
img_path=os.path.join(dir_path,name,img)
img_data=cv2.imread(img_path,cv2.IMREAD_COLOR)
shape=(img_data.shape[1],img_data.shape[0])
#img_data=cv2.resize(img_data,(640,480))
h_n = int(80 * np.ceil(img_data.shape[0] / 80))
w_n = int(80 * np.ceil(img_data.shape[1] / 80))
normed = np.zeros((h_n, w_n, 3), dtype=np.float32) / 255.
normed[0:img_data.shape[0], 0:img_data.shape[1], :] = img_data
plt.imshow(img_data)
plt.draw()
plt.show()
#print(img_data.shape)
#print(img_data)
sample_list, scaled = sampler.sample_and_filter(img_data)
obj = {'origin': utils.encode_img(img_data), 'scaled': utils.encode_img(scaled),'shape':shape}
pickle.dump(obj, fp, -1)
for sample in sample_list:
pickle.dump(sample, fp, -1)
cnt = cnt + 1
self.logger.info(cnt)
self.count = self.count + 1
fp.close()
def root():
print('the request:', request)
x = request.get_json()
img = x['input']
img = decode_img(img)
predictions = model(img)
img = visualise_keypoints(img, predictions)
img = encode_img(img)
return json.dumps({
'statusCode': 200,
'response': img,
'headers': {
'Access-Control-Allow-Origin': '*'
}
})
def sample_thread(sock, capture_pool):
sampler = TestSampler()
logger = logging.getLogger('base')
cnt = 0
while cnt < 20:
with capture_pool.pool_lock:
if capture_pool.pool:
img = capture_pool.pool.pop()
sample_list, scaled = sampler.sample_and_filter(img)
obj = {
'img': utils.encode_img(scaled),
'shape': img.shape,
'sample': False
}
network.send_obj(sock, obj)
for sample in sample_list:
network.send_obj(sock, sample)
cnt = cnt + 1
logger.info(cnt)
else:
time.sleep(0.02)
def start_client_with_file(ip_addr, port, file_path):
"""
利用pkl文件来模拟服务端的数据采集
:param ip_addr: 服务端的ip地址
:param port: 服务器端口
:param file_path: 待使用的pkl文件位置
"""
logger = logging.getLogger('base')
logger.info('start_client')
fp = open(file_path, 'rb')
conn = connection(ip_addr, port)
conn.start_connection()
cnt = 0
sampler = dataprocess.Sampler.TestSampler()
start = time.time()
while True:
try:
data = pickle.load(fp)
data = utils.decode_jpeg(data)
sample_list, scaled = sampler.sample_and_filter(data)
obj = {
'img': utils.encode_img(scaled),
'shape': data.shape,
'sample': False
}
network.send_obj(conn.socket, obj)
for sample in sample_list:
network.send_obj(conn.socket, sample)
# TODO:考虑还有没有优化的策略
cnt = cnt + 1
if cnt % 10 == 0:
curr_time = time.time()
logger.info("sent {} frames in {} sec,fps:{}".format(
cnt, curr_time - start, cnt / (curr_time - start)))
except EOFError:
break
conn.stop_connection()
fp.close()
def write_predict(raw_image, graph, fps):
with tf.Session(graph=graph) as sess:
# Encode test image
raw_img, test_input = encode_img(raw_image, MODEL_SIZE)
print('test_input shape', test_input.shape)
# Run tf model
pred = sess.run(y, feed_dict={x: test_input})
# Handle model output
boxes, scores, classes, nums = output_boxes( \
pred, MODEL_SIZE,
max_output_size=MAX_OUTPUT_SIZE,
max_output_size_per_class=MAX_OUTPUT_SIZE_PER_CLASS,
iou_threshold=IOU_THRESHOLD,
confidence_threshold=CONFIDENCE_THRESHOLD)
img = draw_outputs(raw_img, boxes, scores, classes, nums, class_names)
# Add fps value
words_color = (0, 0, 255) #BGR
if fps is not None:
cv2.putText(img, "FPS: {:.2f}".format(fps), (20, 40),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, words_color, 1)
# Write final result
cv2.imwrite('result.jpg', img)
print('scores', scores.eval())
def start_capture_no_bolck(self,sock):
self.srhandler=SRHandler(sock,self.sampler)
self.srhandler.setDaemon(True)
self.srhandler.start()#开启接收线程
with picamera.PiCamera() as camera:
rawFrame = PiRGBArray(camera, (640, 480))
camera.resolution = (640, 480)
camera.framerate = 15 #目前能够到达17fps左右的采集速度，推测此前瓶颈在于网络传输
shape=(640,480)
time.sleep(2)
self.start = time.time()
cnt=0
start=self.start
for frame in camera.capture_continuous(rawFrame, 'rgb', use_video_port=True):
frame=np.array(frame.array, dtype=np.uint8)
sample_list, scaled = self.sampler.sample_and_filter(frame)
obj={'img':utils.encode_img(scaled),'shape':shape,'sample':False}
network.send_obj(sock,obj)
for sample in sample_list:
network.send_obj(sock, sample)
#TODO:考虑还有没有优化的策略
cnt = cnt + 1
if cnt%10==0:
curr_time=time.time()
self.logger.info("sent {} frames in {} sec,fps:{}".format(cnt, curr_time - start,
cnt / (curr_time - start)))
if cnt>200:
self.logger.info('process end')
break
#self.logger.info(cnt)
self.count = self.count + 1
rawFrame.seek(0)
rawFrame.truncate()
if self.terminate:
self.logger.info('terminate')
break
import requests
from PIL import Image
import json
from io import BytesIO
import base64
from utils import encode_img, decode_img
from time import time
host = '18.133.254.35'
host = 'localhost'
img = Image.open('Images/walking.jpg')
img = encode_img(img)
s = {
'input': img
}
r = requests.post(f'http://{host}:5000/', data=json.dumps(s), headers={'content-type': 'application/json'})
print(r)
print(r.json())
r = r.json()
img = decode_img(r['response'])
img.show()
def get_id(image):
return encode_img("db/" + image + ".jpg")
def __make_base64_encoded_string(self, img_path):
encoded_img = utils.encode_img(img_path)
return utils.process_base64(encoded_img)
