def do_POST(self):
length = int(self.headers['content-length'])
print length
if length > 10000000:
print "file to big"
read = 0
while read < length:
read += len(self.rfile.read(min(66556, length - read)))
self.respond("file to big")
return
else:
form = cgi.FieldStorage(
fp=self.rfile,
headers=self.headers,
environ={'REQUEST_METHOD':'POST',
'CONTENT_TYPE':self.headers['Content-Type'],
})
data = form['file'].file.read()
open("test.jpg", "wb").write(data)
#call facedetect
facedetect.facedetect()
#show result
img = open("result.png", "rb")
self.imgrespond(img.read())
img.close()
def __init__(self,modelfile):
self.fd_detector=facedetect.facedetect().get_instance()
cudnn.enabled = True
self.gpu = "cuda:0"
self.model = importlib.import_module("resnet").inference()
saved_state_dict = torch.load(modelfile)
self.model.load_state_dict(saved_state_dict)
self.transformations = transforms.Compose([transforms.Scale(224),
transforms.CenterCrop(224), transforms.ToTensor(),
transforms.Normalize(mean=[0.4, 0.4, 0.4], std=[0.2, 0.2, 0.2])])
self.model.cuda(self.gpu)
self.model.eval() # Change model to 'eval' mode (BN uses moving mean/var).
def facedetectapi():
if request.method == 'POST':
f = request.files['image']
basepath = os.path.dirname(__file__)
upload_path = os.path.join(basepath, 'postimg.jpg')
f.save(upload_path)
img = cv2.imread(upload_path) # 读取从终端上传的图像
result_dict = facedetect(img) # 对读取到的图像进行人脸检测
# print(result_Dict)
# result_trans = [ str(x) for x in result_list ]
result_json = json.dumps(result_dict)
return result_json + '\n'
return 'Error Format'
def demo(args):
face_fd=facedetect().get_instance("tf-ssh")
cam=cv2.VideoCapture(0)
cam.set(3,1280)
cam.set(4,720)
while(1):
_,img=cam.read()
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
rects, imgs=face_fd.findfaces(img)
if len(rects)==0:
print ("find no face")
else:
for index , (rect, faceimg) in enumerate( zip(rects,imgs) ) :
cv2.imshow("1",faceimg)
cv2.waitKey(1)
return args
def img_preprocess(img):
processimg = cv2.resize(img, (96, 96))
processimg = processimg.astype(np.float32)
processimg = np.transpose(processimg, (2, 0, 1))
processimg = np.expand_dims(processimg, 0)
processimg = processimg / 255.0
processimg = (processimg - 0.5) / 0.5
return processimg
if __name__ == '__main__':
args = parse_args()
fd_detector = facedetect.facedetect("tf-ssh")
cudnn.enabled = True
gpu = "cuda:0"
snapshot_path = args.snapshot
model = resnet.inference(10).cuda()
saved_state_dict = torch.load(snapshot_path)
model.load_state_dict(saved_state_dict)
model.eval()
camera = cv2.VideoCapture(0)
while True:
import os
import sys
sys.path.append("/home/hanson/pytools/lib")
import faceutils as fu
import importlib
import cv2
from facedetect import facedetect
module = importlib.import_module("WFLW106" + "generateBBox")
face_fd = facedetect().get_instance("tf-ssh")
for imgpath, label_rect, label_landmark5p, label_landmark106p in module.generateBBox(
):
image = cv2.imread(imgpath)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
facerects, _ = face_fd.findfaces(image)
target_rect_tmp = max(facerects, key=lambda x: fu.IOU(label_rect, x))
if fu.IOU(target_rect_tmp, label_rect) > 0.65:
target_rect = target_rect_tmp
else:
target_rect = label_rect
roi_image = target_rect.get_roi(image)
scale_landmark = target_rect.projectlandmark(label_landmark5p, scale=0)
#fu.showimg(roi_image, faceboxs = [target_rect], landmarks = [label_landmark])
#fu.showimg(roi_image, landmarks = [scale_landmark] ,landmarkscale=0)
for i in scale_landmark:
import numpy as np
import cv2
# from facebox import Face2point
import tensorflow as tf
import tensorflow.contrib.slim as slim
import utils
def softmax(x):
x_exp = np.exp(x)
x_sum = np.sum(x_exp, axis=0, keepdims=True)
s = x_exp / x_sum
return s
face_fd = facedetect("tf-ssh")
ckpt = tf.train.latest_checkpoint(
"/home/hanson/work/FaceHeadpose_TF/saved_models/20190801-145743/models")
meta = ckpt + ".meta"
print ckpt
saver = tf.train.import_meta_graph(meta)
input_tensor = tf.get_default_graph().get_tensor_by_name("input:0")
yaw_output_tensor = tf.get_default_graph().get_tensor_by_name(
"vgg_16/yaw_fc8/BiasAdd:0")
pitch_output_tensor = tf.get_default_graph().get_tensor_by_name(
"vgg_16/pitch_fc8/BiasAdd:0")
roll_output_tensor = tf.get_default_graph().get_tensor_by_name(
"vgg_16/roll_fc8/BiasAdd:0")
return args
def img_preprocess(img):
processimg = cv2.resize(img, (112, 112))
processimg = processimg.astype(np.float32)
processimg = np.transpose(processimg, (2, 0, 1))
processimg = np.expand_dims(processimg, 0)
processimg = processimg / 255.0
processimg = (processimg - 0.4) / 0.2
return processimg
if __name__ == '__main__':
args = parse_args()
fd_detector = facedetect.facedetect().get_instance()
cudnn.enabled = True
gpu = "cuda:0"
snapshot_path = args.snapshot
model = importlib.import_module("resnet").inference()
saved_state_dict = torch.load(snapshot_path)
model.load_state_dict(saved_state_dict)
transformations = transforms.Compose([
transforms.Resize(112),
transforms.CenterCrop(112),
transforms.ToTensor(),
transforms.Normalize(mean=[0.4, 0.4, 0.4], std=[0.2, 0.2, 0.2])
def generatePic(target_name):
target_image_dir = target_name + "/images"
target_106p_label_file = target_name + "/landmark106p_label.txt"
target_5p_label_file = target_name + "/landmark5p_label.txt"
target_6p_label_file = target_name + "/landmark6p_label.txt"
if not os.path.exists(target_image_dir):
os.makedirs(target_image_dir)
module = importlib.import_module(target_name + "generateBBox")
face_fd = facedetect("tf-ssh")
label5p_f = open(target_5p_label_file, "w")
label6p_f = open(target_6p_label_file, "w")
label106p_f = open(target_106p_label_file, "w")
img_cnt = 0
for imgpath, label_rect, label_landmark5p, label_landmark6p, label_landmark106p in module.generateBBox(
):
if label_rect.width < 40 or label_rect.height < 40:
continue
print(imgpath)
image = cv2.imread(imgpath)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
facerects, _ = face_fd.findfaces(image)
target_rect = label_rect
if len(facerects) > 0:
target_rect_tmp = max(facerects,
key=lambda x: fu.IOU(label_rect, x))
if fu.IOU(target_rect_tmp, label_rect) > 0.65:
target_rect = target_rect_tmp
roi_image = target_rect.get_roi(image)
roi_image = cv2.cvtColor(roi_image, cv2.COLOR_RGB2BGR)
scale_landmark5p = target_rect.projectlandmark(label_landmark5p,
scale=0)
scale_landmark6p = target_rect.projectlandmark(label_landmark6p,
scale=0)
scale_landmark106p = target_rect.projectlandmark(label_landmark106p,
scale=0)
cv2.imwrite("%s/%d.jpg" % (target_image_dir, img_cnt), roi_image)
label5p_f.write("%d.jpg" % img_cnt)
for point in scale_landmark5p:
label5p_f.write(" %d %d" % (point[0], point[1]))
label5p_f.write("\n")
label6p_f.write("%d.jpg" % img_cnt)
for point in scale_landmark6p:
label6p_f.write(" %d %d" % (point[0], point[1]))
label6p_f.write("\n")
label106p_f.write("%d.jpg" % img_cnt)
for point in scale_landmark106p:
label106p_f.write(" %d %d" % (point[0], point[1]))
label106p_f.write("\n")
img_cnt += 1
label5p_f.close()
label6p_f.close()
label106p_f.close()