def train_hs512(lr=1e-4,
freeze_bn=False,
optim=None,
batch_size=8,
weights_path=None,
save_weights_only=True,
epochs=25):
if weights_path is None:
weights_path = 'VGG_VOC0712Plus_SSD_512x512_ft_iter_160000.h5'
img_height = 512 # 1088 // 2 # Height of the input images
img_width = 512 # 2048 // 2 # Width of the input images
img_channels = 3 # Number of color channels of the input images
# subtract_mean = [104, 117, 123] # The per-channel mean of the images in the dataset
subtract_mean = [138, 138,
138] # The per-channel mean of the images in the dataset
swap_channels = False
n_classes = 20 # The number of positive classes, e.g. 20 for Pascal VOC, 80 for MS COCO.
scales = [0.04, 0.1, 0.26, 0.42, 0.58, 0.74, 0.9, 1.06] # MS COCO scales
# scales = [0.07, 0.15, 0.3, 0.45, 0.6, 0.75, 0.9, 1.05]
aspect_ratios = [
[1.0,
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
], [1.0, 2.0, 0.5]
] # The anchor box aspect ratios used in the original SSD300; the order matters
two_boxes_for_ar1 = True
# The space between two adjacent anchor box center points for each predictor layer.
steps = [8, 16, 32, 64, 128, 256, 512]
offsets = [0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5]
limit_boxes = False # Whether or not you want to limit the anchor boxes to lie entirely within the image boundaries
variances = [0.1, 0.1, 0.2, 0.2]
coords = 'centroids'
normalize_coords = True
# 1: Build the Keras model
K.clear_session()
model, pred_sizes = ssd_512(image_size=(img_height, img_width,
img_channels),
n_classes=n_classes,
l2_regularization=0.0005,
scales=scales,
aspect_ratios_per_layer=aspect_ratios,
two_boxes_for_ar1=two_boxes_for_ar1,
steps=steps,
offsets=offsets,
limit_boxes=limit_boxes,
variances=variances,
coords=coords,
normalize_coords=normalize_coords,
subtract_mean=None,
divide_by_stddev=None,
swap_channels=swap_channels,
return_predictor_sizes=True)
# 2: Load the trained VGG-16 weights into the model.
model.load_weights(weights_path, by_name=True)
# 3: Instantiate the optimizer and the SSD loss function and compile the model
if optim is None:
optim = SGD(lr=lr, momentum=0.9)
ssd_loss = SSDLoss(neg_pos_ratio=3, n_neg_min=0, alpha=1.0)
# 4: freeze the base model if needed
if freeze_bn:
for l in model.layers[:38]:
l.trainable = False
# 5: compile model
model.compile(optimizer=optim, loss=ssd_loss.compute_loss)
## Prepare data generation
# 1: Instantiate to `BatchGenerator` objects: One for training, one for validation.
train_dataset = BatchGenerator(
box_output_format=['class_id', 'xmin', 'ymin', 'xmax', 'ymax'])
val_dataset = BatchGenerator(
box_output_format=['class_id', 'xmin', 'ymin', 'xmax', 'ymax'])
# 2: Parse the image and label lists for the training and validation datasets. This can take a while.
images_path_root = './Datasets/'
train_combined_labels = './Datasets/train_combined_ssd_512.txt'
val_labels = './Datasets/val_ssd_512.txt'
train_dataset.parse_csv(images_dir=images_path_root,
labels_filename=train_combined_labels,
input_format=[
'image_name', 'class_id', 'xmin', 'xmax',
'ymin', 'ymax'
])
val_dataset.parse_csv(images_dir=images_path_root,
labels_filename=val_labels,
input_format=[
'image_name', 'class_id', 'xmin', 'xmax', 'ymin',
'ymax'
])
# 3: Instantiate an encoder that can encode ground truth labels into the format needed by the SSD loss function.
ssd_box_encoder = SSDBoxEncoder(img_height=img_height,
img_width=img_width,
n_classes=n_classes,
predictor_sizes=pred_sizes,
min_scale=None,
max_scale=None,
scales=scales,
aspect_ratios_global=None,
aspect_ratios_per_layer=aspect_ratios,
two_boxes_for_ar1=two_boxes_for_ar1,
steps=steps,
offsets=offsets,
limit_boxes=limit_boxes,
variances=variances,
pos_iou_threshold=0.5,
neg_iou_threshold=0.2,
coords=coords,
normalize_coords=normalize_coords)
# 4: Set the image processing / data augmentation options and create generator handles.
train_generator = train_dataset.generate(
batch_size=batch_size,
shuffle=True,
train=True,
ssd_box_encoder=ssd_box_encoder,
equalize=False,
brightness=(0.5, 2, 0.5),
flip=0.5,
translate=False,
scale=False,
max_crop_and_resize=(img_height, img_width, 1, 3),
# This one is important because the Pascal VOC images vary in size
random_pad_and_resize=(img_height, img_width, 1, 3, 0.5),
# This one is important because the Pascal VOC images vary in size
random_crop=False,
crop=False,
resize=False,
gray=True,
limit_boxes=True,
# While the anchor boxes are not being clipped, the ground truth boxes should be
include_thresh=0.4)
val_generator = val_dataset.generate(
batch_size=batch_size,
shuffle=True,
train=True,
ssd_box_encoder=ssd_box_encoder,
equalize=False,
brightness=False,
flip=False,
translate=False,
scale=False,
max_crop_and_resize=(img_height, img_width, 1, 3),
# This one is important because the Pascal VOC images vary in size
random_pad_and_resize=(img_height, img_width, 1, 3, 0.5),
# This one is important because the Pascal VOC images vary in size
random_crop=False,
crop=False,
resize=False,
gray=True,
limit_boxes=True,
include_thresh=0.4)
# Get the number of samples in the training and validations datasets to compute the epoch lengths below.
n_train_samples = train_dataset.get_n_samples()
n_val_samples = val_dataset.get_n_samples()
# ## 4. Run the training
fingerprint = 'ssd{:%Y-%m-%d_%H-%M-%S}'.format(datetime.datetime.now())
tbCallBack = TensorBoard(log_dir='./Graph/{}'.format(fingerprint),
histogram_freq=0,
batch_size=batch_size,
write_graph=True)
checkpointer = ModelCheckpoint(
'./saved/{{val_loss:.2f}}__{}_best_weights.h5'.format(fingerprint),
monitor='val_loss',
verbose=1,
save_best_only=save_weights_only,
save_weights_only=True,
mode='auto',
period=1)
learning_rate_reduction = ReduceLROnPlateau(monitor='val_loss',
patience=5,
verbose=1,
factor=0.5,
min_lr=1e-6)
stopper = EarlyStopping(monitor='val_loss', min_delta=0.001, patience=10)
epochs = 30
history = model.fit_generator(
generator=train_generator,
steps_per_epoch=ceil(n_train_samples / batch_size),
epochs=epochs,
callbacks=[checkpointer, learning_rate_reduction, stopper, tbCallBack],
validation_data=val_generator,
validation_steps=ceil(n_val_samples / batch_size))
dense_layer_size = 512
# Normalization parameters.
mean = 0.408808
std = 0.237583
t_mean = -0.041212
t_std = 0.323931
p_mean = -0.000276
p_std = 0.540958
# Models.
head_detector = ssd_512(image_size=(in_size_detector, in_size_detector, 3), n_classes=1, min_scale=0.1, max_scale=1, mode='inference')
head_detector.load_weights(detector_path)
pose_estimator = mpatacchiola_generic(in_size_estimator, num_conv_blocks, num_filters_start, num_dense_layers, dense_layer_size)
pose_estimator.load_weights(estimator_path)
# Get video source.
video_source = input("Input stream: ")
# If video source is a device convert the string identifying it to integer.
try:
video_source = int(video_source)
except:
pass
def main():
'''
This function acts as a testbench for the function clean_pointing04, using it to perform the basic processing of the
Pointing'04 dataset from a set of default values defined below.
'''
# Source paths.
pointing04_dir = 'original/HeadPoseImageDatabase/'
# Destination path.
destination_dir = 'clean/pointing04/'
# Detector model path.
head_detector_path = 'models/head-detector.h5'
# Detection parameters.
in_size = 512
out_size = 64
confidence_threshold = 0.75
# Output parameters.
grayscale_output = True
downscaling_interpolation = cv2.INTER_LINEAR
# Number of splits for class assignation.
num_splits_tilt = 8
num_splits_pan = 8
# Ratios for train/test and train/validation split.
test_ratio = 0.2
validation_ratio = 0.2
# Detector model.
detector = ssd_512(image_size=(in_size, in_size, 3),
n_classes=1,
min_scale=0.1,
max_scale=1,
mode='inference')
detector.load_weights(head_detector_path)
# Check if output directory exists.
try:
os.mkdir(destination_dir)
print("Directory", destination_dir, "created.")
except FileExistsError:
print("Directory", destination_dir, "already exists.")
shutil.rmtree(destination_dir)
os.mkdir(destination_dir)
# Actual cleaning.
clean_pointing04(pointing04_dir, destination_dir, detector,
confidence_threshold, out_size, grayscale_output,
downscaling_interpolation)
# Assign classes.
class_assign(destination_dir, num_splits_tilt, num_splits_pan)
# Split dataset.
split_dataset(destination_dir, test_ratio, validation_ratio)
# Get normalization parameters.
find_norm_parameters(destination_dir)
# OPTIONAL: Save dataset as numpy arrays (for uploading to Google Colab).
store_dataset_arrays(destination_dir)
clip_boxes = False # Whether or not to clip the anchor boxes to lie entirely within the image boundaries
variances = [
0.1, 0.1, 0.2, 0.2
] # The variances by which the encoded target coordinates are divided as in the original implementation
normalize_coords = True
# 1: Build the Keras model
K.clear_session() # Clear previous models from memory.
model = ssd_512(image_size=(img_height, img_width, img_channels),
n_classes=n_classes,
mode='training',
l2_regularization=0.0005,
scales=scales,
aspect_ratios_per_layer=aspect_ratios,
two_boxes_for_ar1=True,
steps=steps,
offsets=offsets,
clip_boxes=False,
variances=variances,
normalize_coords=normalize_coords,
subtract_mean=mean_color,
swap_channels=swap_channels)
# TODO: Scales ?
# TODO: aspect_ratios_per_layer ?
# TODO: two_boxes_for_ar1 ?
# TODO: steps ?
# TODO: offsets ?
# TODO: clip_boxes ?
# TODO: variances ?
# TODO: normalize_coords ?
# TODO: subtract_mean ?
def main():
rospy.init_node('new_detect_pkg', anonymous=True)
ic = image_converter()
icp = image_converter_pointcloud()
r = rospy.Rate(50)
rospy.sleep(0.5)
publisher()
print('---------initialization model...please wait----------')
# ssd_entity = SSD_entity()
img_height = 512 # 272 Height of the input images
img_width = 512 # 480 Width of the input images
img_channels = 3 # Number of color channels of the input images
# 1: Build the Keras model
K.clear_session() # Clear previous models from memory.
model = ssd_512(
image_size=(img_height, img_width, 3),
n_classes=20,
mode='inference',
l2_regularization=0.0005,
scales=[
0.07, 0.15, 0.3, 0.45, 0.6, 0.75, 0.9, 1.05
], # The scales for MS COCO are [0.04, 0.1, 0.26, 0.42, 0.58, 0.74, 0.9, 1.06]
aspect_ratios_per_layer=[[1.0, 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], [1.0, 2.0, 0.5]],
two_boxes_for_ar1=True,
steps=[8, 16, 32, 64, 128, 256, 512],
offsets=[0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5],
clip_boxes=False,
variances=[0.1, 0.1, 0.2, 0.2],
normalize_coords=True,
subtract_mean=[123, 117, 104],
swap_channels=[2, 1, 0],
confidence_thresh=0.5,
iou_threshold=0.45,
top_k=200,
nms_max_output_size=400)
print("Model built.")
# 2: Load the sub-sampled weights into the model.
# Load the weights that we've just created via sub-sampling.
# model.load_weights('/home/ogai1234/catkin_ws/src/detect_pkg/bin/ssd300_weights_epoch-45_loss-4.3010_val_loss-4.3788.h5', by_name=True)
# model.load_weights('/home/ogai1234/catkin_ws/src/detect_pkg/bin/VGG_VOC0712Plus_SSD_300x300_ft_iter_160000.h5', by_name=True)
model.load_weights(
'/home/ogai1234/catkin_ws/src/detect_pkg/bin/VGG_VOC0712Plus_SSD_512x512_ft_iter_160000.h5',
by_name=True)
print('---------model done----------')
# model.load_weights(weights_path, by_name=True)
# print("Weights file loaded:", weights_path)
# classes = ["background", "dog", "umbrellaman", "cone", "car", "bicycle", "person"]
classes = [
'background', 'Aeroplane', 'Bicycle', 'Bird', 'Boat', 'Bottle', 'Bus',
'Car', 'Cat', 'Chair', 'Cow', 'Diningtable', 'Dog', 'Horse',
'Motorbike', 'Person', 'Pottedplant', 'Sheep', 'Sofa', 'Train',
'Tvmonitor'
]
key = ''
t0 = time.time()
frame_code = 1
linewidth = 3
figsize = (10, 10)
# initilize the filter cach
global record_xmin, record_xmax, record_ymin, record_ymax, record_cls_id, record_obj_id
global record_obj_num_appeared, total_objPerFrame
record_obj_num_appeared = 0
record_xmin = [[0 for i in range(11)] for i in range(2)]
record_xmax = [[0 for i in range(11)] for i in range(2)]
record_ymin = [[0 for i in range(11)] for i in range(2)]
record_ymax = [[0 for i in range(11)] for i in range(2)]
record_cls_id = [[0 for i in range(11)] for i in range(2)]
record_obj_id = [[0 for i in range(11)] for i in range(2)]
global objPerFrame
objPerFrame = 0
total_objPerFrame = 0
font = cv2.FONT_HERSHEY_TRIPLEX
while (key != 113) and (not rospy.is_shutdown()):
t1 = time.time()
#one frame start
frame_code = frame_code + 1
print("_________________________Frame:", frame_code)
# for ros topic publish
global objPerlastFrame, type_code_list, confidence_list, distance_list, x_list, y_list, z_list
objPerlastFrame = total_objPerFrame
# calculate the num of object appeared in one frame
objPerFrame = 0
total_objPerFrame = 0
# Ros message content initial
type_code_list = [0 for i in range(11)]
confidence_list = [0 for i in range(11)]
distance_list = [0 for i in range(11)]
x_list = [0 for i in range(11)]
y_list = [0 for i in range(11)]
z_list = [0 for i in range(11)]
# load zed image from ros topic
image = ic.zed_image # image.shape = (720, 1280, 3)
frame = image
frame_old = frame
# frame = frame[:,:,0:3]
frame = cv2.resize(frame, (img_width, img_height))
frame_np = np.array(frame)
frame_np = frame_np[np.newaxis, :, :, :]
# put frame into SSD network to do classification
y_pred = model.predict(frame_np)
confidence_threshold = 0.5
y_pred_thresh = [
y_pred[k][y_pred[k, :, 1] > confidence_threshold]
for k in range(y_pred.shape[0])
]
colors = dict()
for box in y_pred_thresh[0]:
if objPerFrame > 9:
break
cls_id = int(box[0])
if cls_id not in colors:
colors[cls_id] = (random.random(), random.random(),
random.random())
# for person and car/train/bus, collect its information and transfer to ROS
if cls_id == 15: #Person
score = box[1]
xmin = int(box[2] * frame_old.shape[1] / img_width)
ymin = int(box[3] * frame_old.shape[0] / img_height)
xmax = int(box[4] * frame_old.shape[1] / img_width)
ymax = int(box[5] * frame_old.shape[0] / img_height)
cv2.rectangle(frame_old, (xmin, ymin), (xmax, ymax),
color=colors[cls_id],
thickness=linewidth)
# class_name = str(cls_id)
# calculate distance and position
x_center = round((xmin + xmax) / 2)
y_center = round((ymin + ymax) / 2)
icp.dict_1 = x_center
icp.dict_2 = y_center
rospy.sleep(0.04)
point_cloud = icp.zed_image_pointcloud
x, y, z = 0, 0, 0
#x,y,z is the position obtained from pointcloud2
for p in point_cloud:
x, y, z = p
break
distance = math.sqrt(x * x + y * y + z * z)
type_code = cls_id
class_name = classes[cls_id]
# Person's typecode is 1
type_code = 1
total_objPerFrame = total_objPerFrame + 1
appeared = frameFilter(cls_id, xmin, xmax, ymin, ymax)
if (appeared == 1):
objRecord2Ros(class_name, score, cls_id, type_code,
distance, x, y, z)
cv2.putText(frame_old,
'{:s} | {:.2f} | {:.2f} |{:}'.format(
class_name, score, distance,
record_obj_id[1][total_objPerFrame]),
(xmin, ymin + 2),
font,
0.5, (255, 255, 255),
thickness=1)
if cls_id == 7: #Car
score = box[1]
xmin = int(box[2] * frame_old.shape[1] / img_width)
ymin = int(box[3] * frame_old.shape[0] / img_height)
xmax = int(box[4] * frame_old.shape[1] / img_width)
ymax = int(box[5] * frame_old.shape[0] / img_height)
cv2.rectangle(frame_old, (xmin, ymin), (xmax, ymax),
color=colors[cls_id],
thickness=linewidth)
# class_name = str(cls_id)
# calculate distance and position
x_center = round((xmin + xmax) / 2)
y_center = round((ymin + ymax) / 2)
icp.dict_1 = x_center
icp.dict_2 = y_center
rospy.sleep(0.04)
point_cloud = icp.zed_image_pointcloud
x, y, z = 0, 0, 0
#x,y,z is the position obtained from pointcloud2
for p in point_cloud:
x, y, z = p
break
distance = math.sqrt(x * x + y * y + z * z)
type_code = cls_id
class_name = classes[cls_id]
# Car's typecode is 4
type_code = 2
total_objPerFrame = total_objPerFrame + 1
appeared = frameFilter(cls_id, xmin, xmax, ymin, ymax)
if (appeared == 1):
objRecord2Ros(class_name, score, cls_id, type_code,
distance, x, y, z)
cv2.putText(frame_old,
'{:s} | {:.2f} |{:}'.format(
class_name, distance,
record_obj_id[1][total_objPerFrame]),
(xmin, ymin + 2),
font,
0.5, (255, 255, 255),
thickness=1)
if cls_id == 6: #Bus
score = box[1]
xmin = int(box[2] * frame_old.shape[1] / img_width)
ymin = int(box[3] * frame_old.shape[0] / img_height)
xmax = int(box[4] * frame_old.shape[1] / img_width)
ymax = int(box[5] * frame_old.shape[0] / img_height)
cv2.rectangle(frame_old, (xmin, ymin), (xmax, ymax),
color=colors[cls_id],
thickness=linewidth)
# class_name = str(cls_id)
# calculate distance and position
x_center = round((xmin + xmax) / 2)
y_center = round((ymin + ymax) / 2)
icp.dict_1 = x_center
icp.dict_2 = y_center
rospy.sleep(0.04)
point_cloud = icp.zed_image_pointcloud
x, y, z = 0, 0, 0
#x,y,z is the position obtained from pointcloud2
for p in point_cloud:
x, y, z = p
break
distance = math.sqrt(x * x + y * y + z * z)
type_code = cls_id
class_name = classes[cls_id]
# Car's typecode is 4
type_code = 2
total_objPerFrame = total_objPerFrame + 1
appeared = frameFilter(cls_id, xmin, xmax, ymin, ymax)
if (appeared == 1):
objRecord2Ros(class_name, score, cls_id, type_code,
distance, x, y, z)
cv2.putText(frame_old,
'{:s} | {:.2f} |{:}'.format(
class_name, distance,
record_obj_id[1][total_objPerFrame]),
(xmin, ymin + 2),
font,
0.5, (255, 255, 255),
thickness=1)
if cls_id == 19: #Train
score = box[1]
xmin = int(box[2] * frame_old.shape[1] / img_width)
ymin = int(box[3] * frame_old.shape[0] / img_height)
xmax = int(box[4] * frame_old.shape[1] / img_width)
ymax = int(box[5] * frame_old.shape[0] / img_height)
cv2.rectangle(frame_old, (xmin, ymin), (xmax, ymax),
color=colors[cls_id],
thickness=linewidth)
# class_name = str(cls_id)
# calculate distance and position
x_center = round((xmin + xmax) / 2)
y_center = round((ymin + ymax) / 2)
icp.dict_1 = x_center
icp.dict_2 = y_center
rospy.sleep(0.04)
point_cloud = icp.zed_image_pointcloud
x, y, z = 0, 0, 0
#x,y,z is the position obtained from pointcloud2
for p in point_cloud:
x, y, z = p
break
distance = math.sqrt(x * x + y * y + z * z)
type_code = cls_id
class_name = classes[cls_id]
# Car's typecode is 4
type_code = 4
total_objPerFrame = total_objPerFrame + 1
appeared = frameFilter(cls_id, xmin, xmax, ymin, ymax)
if (appeared == 1):
objRecord2Ros(class_name, score, cls_id, type_code,
distance, x, y, z)
cv2.putText(frame_old,
'{:s} | {:.2f} |{:}'.format(
class_name, distance,
record_obj_id[1][total_objPerFrame]),
(xmin, ymin + 2),
font,
0.5, (255, 255, 255),
thickness=1)
# if want show all classes, uncomment this line , and comment the upper 2 ifs
# objRecord2Ros(class_name,score,cls_id,type_code,distance,x,y,z)
# cv2.putText(frame, '{:s} | {:.2f} |{:}'.format(class_name,distance,record_obj_id[1][total_objPerFrame]), (xmin, ymin+2), font, 0.5, (255, 255, 255), thickness=1)
# draw the bounding box
# print format: class|conf|distance|x|y
# show the image with bounding box
# image_back = cv2.resize(frame, (1080,720))
cv2.imshow("image_back", frame_old)
key = cv2.waitKey(1)
t21 = time.time()
# calculate fps
print('fps {:f}'.format(1 / (t21 - t1)))
talker()
# last frame info move forward
for i in range(10):
# print('record_xmin[0][i]:',record_xmin[0][i+1])
# print('record_xmin[1][i]:',record_xmin[1][i+1])
record_xmin[0][i + 1] = record_xmin[1][i + 1]
record_xmax[0][i + 1] = record_xmax[1][i + 1]
record_ymin[0][i + 1] = record_ymin[1][i + 1]
record_ymax[0][i + 1] = record_ymax[1][i + 1]
record_cls_id[0][i + 1] = record_cls_id[1][i + 1]
record_obj_id[0][i + 1] = record_obj_id[1][i + 1]
record_xmin[1][i + 1] = 0
record_xmax[1][i + 1] = 0
record_ymin[1][i + 1] = 0
record_ymax[1][i + 1] = 0
record_cls_id[1][i + 1] = 0
record_obj_id[1][i + 1] = 0
if record_obj_num_appeared > 10000:
record_obj_num_appeared = 0
model_dir = 'models/'
model_name = 'head-detector'
matconvnet_model_path = model_dir + model_name + '.mat'
keras_model_path = model_dir + model_name + '.h5'
layer_mapping_path = model_dir + 'layers.csv'
# Model parameters.
img_height = 512
img_width = 512
# Model instance.
model = ssd_512(image_size=(img_height, img_width, 3), n_classes=1, min_scale=0.1, max_scale=1, mode='inference')
# Load weights.
model_ori = loadmat(matconvnet_model_path)
# Load layer mapping between MatConvNet and Keras models.
layers = pd.read_csv(layer_mapping_path)
# For each entry in the mapping, set its weights and biases from the original model:
for index, layer in layers.iterrows():
print("Assigning layer " + layer['name'] + "...")
