def demo_for_human36m():
joint_world = [[-91.679, 154.404, 907.261],
[-223.23566, 163.80551, 890.5342],
[-188.4703, 14.077106, 475.1688],
[-261.84055, 186.55286, 61.438915],
[39.877888, 145.00247, 923.98785],
[-11.675994, 160.89919, 484.39148],
[-51.550297, 220.14624, 35.834396],
[-132.34781, 215.73018, 1128.8396],
[-97.1674, 202.34435, 1383.1466],
[-112.97073, 127.96946, 1477.4457],
[-120.03289, 190.96477, 1573.4],
[25.895456, 192.35947, 1296.1571],
[107.10581, 116.050285, 1040.5062],
[129.8381, -48.024918, 850.94806],
[-230.36955, 203.17923, 1311.9639],
[-315.40536, 164.55284, 1049.1747],
[-350.77136, 43.442127, 831.3473],
[-102.237045, 197.76935, 1304.0605]]
joint_world = np.asarray(joint_world)
# 关节点连接线
kps_lines = ((0, 7), (7, 8), (8, 9), (9, 10), (8, 11), (11, 12), (12, 13),
(8, 14), (14, 15), (15, 16), (0, 1), (1, 2), (2, 3), (0, 4),
(4, 5), (5, 6))
# show in 世界坐标系
vis.vis_3d(joint_world,
kps_lines,
coordinate="WC",
title="WC",
set_lim=True,
isshow=True)
kp_vis = CameraTools()
# show in 相机坐标系
joint_cam = kp_vis.convert_wc_to_cc(joint_world)
vis.vis_3d(joint_cam,
kps_lines,
coordinate="CC",
title="CC",
set_lim=True,
isshow=True)
joint_img = kp_vis.convert_cc_to_ic(joint_cam)
joint_world1 = kp_vis.convert_cc_to_wc(joint_cam)
vis.vis_3d(joint_world1,
kps_lines,
coordinate="WC",
title="WC",
set_lim=True,
isshow=True)
# show in 像素坐标系
kpt_2d = joint_img[:, 0:2]
image_path = "./data/s_01_act_02_subact_01_ca_02_000001.jpg"
image = image_processing.read_image(image_path)
image = image_processing.draw_key_point_in_image(image,
key_points=[kpt_2d],
pointline=kps_lines)
image_processing.cv_show_image("image_dict", image)
def show_landmark_boxes(win_name, image, bboxes, scores, landms):
'''
显示landmark和boxes
:param win_name:
:param image:
:param landmarks_list: [[x1, y1], [x2, y2]]
:param bboxes: [[ x1, y1, x2, y2],[ x1, y1, x2, y2]]
:return:
'''
image = image_processing.draw_landmark(image, landms, vis_id=True)
image = image_processing.draw_image_bboxes_text(image,
bboxes,
scores,
color=(0, 0, 255))
image_processing.cv_show_image(win_name, image, waitKey=1)
def demo_for_kinect():
# data_dir = "E:/git/python-learning-notes/tutorial/kinect2/dataset/kitnect3d"
data_dir = "/media/dm/dm/X2/Pose/dataset/kitnet_data/panjinquan"
image, joint_world = load_data(data_dir, flag=5)
joint_world = -joint_world * 1000
kps_lines = [
[0, 1],
[1, 20],
[20, 2],
[2, 3], # Spine
[20, 4],
[4, 5],
[5, 6],
[6, 7],
[7, 21],
[7, 22], # Left arm and hand
[20, 8],
[8, 9],
[9, 10],
[10, 11],
[11, 23],
[11, 24], # Right arm and hand
[0, 12],
[12, 13],
[13, 14],
[14, 15], # Left leg
[0, 16],
[16, 17],
[17, 18],
[18, 19]
] # Right leg
# show in 世界坐标系
vis.vis_3d(joint_world, kps_lines, coordinate="WC", title="WC")
kp_vis = CameraTools()
# show in 相机坐标系
joint_cam = kp_vis.convert_wc_to_cc(joint_world)
vis.vis_3d(joint_cam, kps_lines, coordinate="CC", title="CC")
joint_img = kp_vis.convert_cc_to_ic(joint_cam)
# show in 像素坐标系
kpt_2d = joint_img[:, 0:2]
image = image_processing.draw_key_point_in_image(image,
key_points=[kpt_2d],
pointline=kps_lines)
image_processing.cv_show_image("image_dict", image)
def show(self, filename, image, keypoints, bboxes, class_name, vis=True):
is_save = True
for i, joints in enumerate(keypoints):
if np.sum(np.asarray(joints[5])) == 0 or np.sum(np.asarray(joints[6])) == 0 or \
np.sum(np.asarray(joints[11])) == 0 or np.sum(np.asarray(joints[12])) == 0:
is_save = False
else:
is_save = True
chest_joint = (np.asarray(joints[5]) + np.asarray(joints[6])) / 2
hip_joint = (np.asarray(joints[11]) + np.asarray(joints[12])) / 2
keypoints[i].append(chest_joint.tolist())
keypoints[i].append(hip_joint.tolist())
if vis:
image_processing.draw_image_bboxes_text(image, bboxes, class_name)
# image_processing.show_image_boxes(None, image, joints_bbox, color=(255, 0, 0))
image = image_processing.draw_key_point_in_image(
image, keypoints, pointline=self.skeleton)
# image_processing.cv_show_image("Det", image, waitKey=0)
# self.save_images(image, filename, is_save)
image_processing.cv_show_image("Det", image)
def label_test(image_dir, filename, class_names=None):
basename = os.path.basename(filename)[:-len('.txt')] + ".jpg"
image_path = os.path.join(image_dir, basename)
image = image_processing.read_image(image_path)
data = file_processing.read_data(filename, split=" ")
label_list, rect_list = file_processing.split_list(data, split_index=1)
label_list = [l[0] for l in label_list]
if class_names:
name_list = file_processing.decode_label(label_list, class_names)
else:
name_list = label_list
show_info = ["id:" + str(n) for n in name_list]
rgb_image = image_processing.show_image_rects_text("object2",
image,
rect_list,
show_info,
color=(0, 0, 255),
drawType="custom",
waitKey=1)
rgb_image = image_processing.resize_image(rgb_image, 900)
image_processing.cv_show_image("object2", rgb_image)
def create_dataset(out_dir, nums, filename, char_set, captcha_height,
captcha_width, captcha_size):
'''
产生样本
:param out_dir: 数据集图片保存目录
:param nums: 产生数据样本个数
:param filename: 保存数据txt文件
:param char_set: 字符数据集
:param captcha_height: 验证码height
:param captcha_width: 验证码width
:param captcha_size: 验证码大小
:return:None
'''
if not os.path.exists(out_dir):
os.mkdir(out_dir)
# 产生一个验证码样本并显示
i = 0
while i < nums:
text, image = gen_captcha_text_and_image(char_set=char_set,
captcha_height=captcha_height,
captcha_width=captcha_width,
captcha_size=captcha_size)
# 产生的验证码图并不一定等于image_height*image_width
if image.shape != (image_height, image_width, 3):
continue
if i == 0:
image_processing.cv_show_image(text, image) # 显示验证码
image_name = str(i) + "_" + text + ".jpg"
image_path = out_dir + "/" + image_name
print(image_path)
image_processing.save_image(image_path, image, toUINT8=False)
text = [c for c in text]
label_list = file_processing.label_encode(text, char_set)
content = [image_name] + label_list
content = ' '.join('%s' % id for id in content)
file_processing.write_data(filename, [content], model='a')
i += 1
def demo_for_kinect():
flip_transform = np.asarray([[1, 0, 0],
[0, -1, 0],
[0, 0, -1]])
# flip_transform = np.linalg.inv(flip_transform)
# data_dir = "E:/git/python-learning-notes/tutorial/kinect2/dataset/kitnect3d"
data_dir = "/media/dm/dm/X2/Pose/dataset/kitnet_data/panjinquan" # flag= 5
# data_dir = "/media/dm/dm/X2/Pose/dataset/kitnet_data/dengjianxiang" # 241,245,348
image, joint_world = load_data(data_dir, flag=503)
h, w, d = image.shape
joint_world = convert_kinect2h36m(joint_world)
joint_world = joint_world * 1000
joint_world = np.dot(-flip_transform, joint_world.T).T # R * (pt - T)
# kps_lines = [[0, 1], [1, 20], [20, 2], [2, 3], # Spine
# [20, 4], [4, 5], [5, 6], [6, 7], [7, 21], [7, 22], # Left arm and hand
# [20, 8], [8, 9], [9, 10], [10, 11], [11, 23], [11, 24], # Right arm and hand
# [0, 12], [12, 13], [13, 14], [14, 15], # Left leg
# [0, 16], [16, 17], [17, 18], [18, 19]] # Right leg
kps_lines = ((0, 7), (7, 8), (8, 9), (9, 10), (8, 11), (11, 12), (12, 13), (8, 14), (14, 15),
(15, 16), (0, 1), (1, 2), (2, 3), (0, 4), (4, 5), (5, 6))
# show in 世界坐标系
vis.vis_3d(joint_world, kps_lines, coordinate="WC", title="WC", set_lim=True)
kp_vis = KeyPointsVisual()
# show in 相机坐标系
joint_cam = kp_vis.convert_wc_to_cc(joint_world)
vis.vis_3d(joint_cam, kps_lines, coordinate="CC", title="CC", set_lim=True)
joint_cam = np.dot(flip_transform, joint_cam.T).T # R * (pt - T)
joint_img = kp_vis.convert_cc_to_ic(joint_cam)
# show in 像素坐标系
kpt_2d = joint_img[:, 0:2]
image = image_processing.draw_key_point_in_image(image, key_points=[kpt_2d], pointline=kps_lines)
image_processing.cv_show_image("image_dict", image)
def demo_for_human36m():
from modules.utils_3d.data import human36m_data
# x,y,z
# joint_world = human36m_data.data0
joint_world = human36m_data.data2 * 1000
# joint_world = human36m_data.data1*1000
joint_world = np.asarray(joint_world)
kps_lines = human36m_data.kps_lines
# show in 世界坐标系
vis.vis_3d(joint_world, kps_lines, coordinate="WC", title="WC", set_lim=True)
kp_vis = KeyPointsVisual()
# show in 相机坐标系
joint_cam = kp_vis.convert_wc_to_cc(joint_world)
vis.vis_3d(joint_cam, kps_lines, coordinate="CC", title="CC", set_lim=True)
joint_img = kp_vis.convert_cc_to_ic(joint_cam)
# show in 像素坐标系
kpt_2d = joint_img[:, 0:2]
image_path = "/media/dm/dm1/git/python-learning-notes/modules/utils_3d/data/s_01_act_02_subact_01_ca_02_000001.jpg"
image = image_processing.read_image(image_path)
image = image_processing.draw_key_point_in_image(image, key_points=[kpt_2d], pointline=kps_lines)
image_processing.cv_show_image("image_dict", image)
def show(self, color_img, align_color_img, depth_img, ir_frame, bboxes):
"""
:param color_img: <class 'tuple'>: (1080, 1920, 4),uint8
:param align_color_img: <class 'tuple'>: (424, 512, 4),uint8
:param depth_img: <class 'tuple'>: (424, 512),uint16
:param ir_frame: <class 'tuple'>: (424, 512, 3),uint8
:return:
"""
# align_color_img = cv2.cvtColor(align_color_img, cv2.COLOR_RGBA2RGB) # 将BGR转为RGB
image_processing.addMouseCallback("depth_img", param=depth_img)
image_processing.addMouseCallback("align_color_img", param=depth_img)
# image_processing.cv_show_image("color_img", color_img, waitKey=1)
image_processing.cv_show_image("align_color_img",
align_color_img,
waitKey=1)
image_processing.cv_show_image("depth_img", depth_img, waitKey=1)
image_processing.cv_show_image("ir_frame", ir_frame, waitKey=1)
image_processing.cv_show_image("ir_frame1", self.ir_frame1, waitKey=1)
self.count += 1
freq = 3
# self.flag = cv2.waitKey(20) & 0xFF == ord('s')
if self.snapshot_dir and len(bboxes) > 0 and self.count % freq == 0:
print("save image:{}".format(self.count))
# self.flag = True
# pre = "{}.png".format(self.count)
prefix = "{}_{}.png".format(self.prefix, self.count)
cv2.imwrite(
os.path.join(self.snapshot_dir, "depth/{}".format(prefix)),
depth_img)
cv2.imwrite(
os.path.join(self.snapshot_dir, "color/{}".format(prefix)),
align_color_img)
cv2.imwrite(
os.path.join(self.snapshot_dir, "ir/{}".format(prefix)),
ir_frame)
cv2.waitKey(20)
self.save_videos(color_img, align_color_img, depth_img, ir_frame,
bboxes)
def deme_test(self, out_voc_ann=None, vis=True):
joints_bbox = []
for lines in self.boxes_label_lists:
# lines = ['lexue_teacher_LHui211_20190921145000_20190921172000_40_000143.jpg', 543.053254438, 527.147928994,
# 456.710059172, 548.733727811, 'person']
image_name = lines[0]
rect = [lines[1], lines[2], lines[3], lines[4]]
label = [lines[5]]
image_path = os.path.join(self.image_dir, image_name)
json_file = image_name[:-len(".jpg")] + ".json"
json_path = os.path.join(self.json_dir, json_file)
if not os.path.exists(image_path):
print("Error:no path: {}".format(json_file))
continue
if not os.path.exists(json_path):
print("Error:no path: {}".format(json_file))
continue
anno = CustomDataset.get_anno(json_path)
if not anno:
print("Error:empty path: {}".format(json_file))
continue
keypoints = self.get_keypoints(anno)
sum = np.sum(np.abs(np.asarray(keypoints)))
if keypoints == [] or sum == 0:
print("Error:empty path: {}".format(json_file))
continue
image = image_processing.read_image(image_path)
if out_voc_ann:
try:
bboxes = image_processing.rects2bboxes([rect])
joints_bbox = self.convert_voc_dataset(
image_name,
image_shape=image.shape,
bboxes=bboxes,
labels=label,
keypoints=keypoints,
out_voc_ann=out_voc_ann)
except Exception as e:
print("Error:empty path: {}".format(json_file))
raise Exception("lines: {}".format(lines))
save_image = True
for i, joints in enumerate(keypoints):
if np.sum(np.asarray(joints[5])) == 0 or np.sum(np.asarray(joints[6])) == 0 or \
np.sum(np.asarray(joints[11])) == 0 or np.sum(np.asarray(joints[12]))==0:
save_image = False
else:
save_image = True
chest_joint = (np.asarray(joints[5]) +
np.asarray(joints[6])) / 2
hip_joint = (np.asarray(joints[11]) +
np.asarray(joints[12])) / 2
keypoints[i].append(chest_joint.tolist())
keypoints[i].append(hip_joint.tolist())
if vis:
image_processing.show_image_rects(None, image, [rect])
# image_processing.show_image_boxes(None, image, joints_bbox, color=(255, 0, 0))
image = image_processing.draw_key_point_in_image(
image, keypoints, pointline=skeleton)
image_processing.cv_show_image("Det", image, waitKey=1)
if save_image:
out_dir = "/media/dm/dm2/project/dataset/COCO/HumanPose/LeXue_teacher/Posture/tmp1"
out_dir = file_processing.create_dir(out_dir)
out_image_path = os.path.join(out_dir, image_name)
image_processing.save_image(out_image_path, image)
face_landmarks = [[landmark[j], landmark[j + 5]] for j in range(5)]
landmarks_list.append(face_landmarks)
landmarks = np.asarray(landmarks_list)
return landmarks
if __name__ == "__main__":
# img = Image.open('some_img.jpg') # modify the image path to yours
# bounding_boxes, landmarks = detect_faces(img) # detect bboxes and landmarks for all faces in the image
# show_results(img, bounding_boxes, landmarks) # visualize the results
# image_path = "/media/dm/dm/project/dataset/face_recognition/NVR/JPEGImages/2000.jpg"
image_path = "/media/dm/dm1/FaceRecognition/torch-Face-Recognize-Pipeline/data/dataset2/zhoujielun/zhoujielun_1.jpg"
image = image_processing.read_image(image_path, colorSpace="RGB")
mt = MTCNN()
bbox_score, landmarks = mt.detect(image)
bboxes, scores, landmarks = mt.adapter_bbox_score_landmarks(bbox_score, landmarks)
# image_processing.show_image_boxes("image",image,bboxes)
# image_processing.show_landmark_boxes("image", image, landmarks, bboxes)
# image_processing.show_landmark_boxes("image2", image, landmarks, bboxes)
faces = image_processing.get_bboxes_image(image, bboxes)
# landmarks2 = mt.landmarks_forward(bbox_score, image)
# bboxes, scores, landmarks2 = mt.adapter_bbox_score_landmarks(bbox_score, landmarks2)
# image_processing.show_landmark_boxes("image2", image, landmarks2, bboxes)
for face in faces:
image_processing.cv_show_image("face", face)
image_processing.show_landmark_boxes("image", image, landmarks, bboxes)
landmarks = mt.face_landmarks_forward([face])
image_processing.show_landmark("landmark", face, landmarks)
# test_data = TorchDataset(filename=test_filename, image_dir=image_dir,repeat=1)
train_loader = DataLoader(dataset=train_data,
batch_size=batch_size,
shuffle=False)
# test_loader = DataLoader(dataset=test_data, batch_size=batch_size,shuffle=False)
# [1]使用epoch方法迭代，TorchDataset的参数repeat=1
for epoch in range(epoch_num):
for step, bach_data in enumerate(train_loader):
batch_image = bach_data['image']
batch_label = bach_data['lable']
image = batch_image[0, :]
image = image.numpy() #image=np.array(image)
image = image.transpose(1, 2, 0) # 通道由[c,h,w]->[h,w,c]
image_processing.cv_show_image("image", image)
print("batch_image.shape:{},batch_label:{}".format(
batch_image.shape, batch_label))
# batch_x, batch_y = Variable(batch_x), Variable(batch_y)
'''
下面两种方式，TorchDataset设置repeat=None可以实现无限循环，退出循环由max_iterate设定
'''
train_data = TorchDataset(filename=train_filename,
image_dir=image_dir,
repeat=None)
train_loader = DataLoader(dataset=train_data,
batch_size=batch_size,
shuffle=False)
# [2]第2种迭代方法
for step, bach_data in enumerate(train_loader):
batch_image = bach_data['image']
alphabet = [
'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n',
'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z'
]
ALPHABET = [
'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N',
'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z'
]
'''设置验证码的信息：
这里验证码字符集只考虑数字的情况,并且验证码大小captcha_size=4
'''
char_set = number
# char_set = number + alphabet + ALPHABET # 如果要预测26个大小写字母，则加上alphabet和ALPHABET字符集
batch_size = 32
image_height = 60
image_width = 160
depth = 3
captcha_size = 4
# 产生一个验证码样本并显示
text, image = create_dataset.gen_captcha_text_and_image(
char_set=char_set,
captcha_height=image_height,
captcha_width=image_width,
captcha_size=captcha_size)
image_processing.cv_show_image(text, image) #显示验证码
print("验证码图像shape:{}".format(image.shape)) # (60, 160, 3)
print("验证码字符个数:{}".format(captcha_size))
# 训练
train(char_set, batch_size, image_height, image_width, depth, captcha_size)
return faces
if __name__ == "__main__":
# img = Image.open('some_img.jpg') # modify the image path to yours
# bounding_boxes, landmarks = detect_faces(img) # detect bboxes and landmarks for all faces in the image
# show_results(img, bounding_boxes, landmarks) # visualize the results
# image_path = "/media/dm/dm/project/dataset/face_recognition/NVR/JPEGImages/2000.jpg"
image_path1 = "/media/dm/dm1/FaceRecognition/torch-Face-Recognize-Pipeline/data/test_images/face1.jpg"
image_path2 = "/media/dm/dm1/FaceRecognition/torch-Face-Recognize-Pipeline/data/test_images/face2.jpg"
face1 = image_processing.read_image(image_path1,
colorSpace="RGB",
resize_height=200,
resize_width=100)
face2 = image_processing.read_image(image_path2,
colorSpace="RGB",
resize_height=100,
resize_width=200)
lmdet = ONetLandmarkDet(device="cuda:0")
# image_processing.cv_show_image("face", face)
# image_processing.show_landmark_boxes("image", image, landmarks, bboxes)
faces = []
faces.append(face1)
faces.append(face2)
landmarks = lmdet.get_faces_landmarks(faces)
alig_faces = lmdet.face_alignment(faces, face_resize=[112, 112])
for i in range(len(faces)):
image_processing.show_landmark("landmark", faces[i], [landmarks[i]])
image_processing.cv_show_image("alig_faces", alig_faces[i])