def create_callbacks(saved_weights_name, tensorboard_logs):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(monitor='val_loss',
min_delta=0.001,
patience=4,
mode='min',
verbose=1)
checkpoint = ModelCheckpoint(saved_weights_name,
monitor='val_loss',
verbose=1,
save_best_only=False,
mode='min',
period=1)
reduce_on_plateau = ReduceLROnPlateau(monitor='val_loss',
factor=0.1,
patience=2,
verbose=1,
mode='min',
epsilon=0.0001,
cooldown=0,
min_lr=0)
tensorboard = TensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
return [early_stop, checkpoint, reduce_on_plateau, tensorboard]
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(monitor='loss',
min_delta=0.01,
patience=7,
mode='min',
verbose=1)
checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=saved_weights_name, # + '{epoch:02d}.h5',
monitor='loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
reduce_on_plateau = ReduceLROnPlateau(monitor='loss',
factor=0.1,
patience=2,
verbose=1,
mode='min',
epsilon=0.01,
cooldown=0,
min_lr=0)
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
return [early_stop, checkpoint, reduce_on_plateau, tensorboard]
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save):
makedirs(tensorboard_logs)
early_stop = EarlyStopping( # early stop
monitor='loss', #based on loss
min_delta=0.01, #if improvment less than 0.01 it's not a real improve
patience=5, #number of times loss didn't improve to stop (5 times try )
mode='min', #based on loss choice we want it min
verbose=1 #print epoch and extra info default is 0
)
checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=saved_weights_name, # + '{epoch:02d}.h5',
monitor='loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
reduce_on_plateau = ReduceLROnPlateau( # reduce learning rate when loss didn't improve
monitor='loss', #based ob loss
factor=0.1, #factir if reduce
patience=2, # every 2 epochs
verbose=1, # 0 equal quit , 1 update
mode='min', #min mode for loss
epsilon=0.01, #change in epsilon
cooldown=
0, #number of epochs to wait before resuming normal operation after lr has been reduced.
min_lr=0 #lower bound on the learning rate.
)
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
return [early_stop, checkpoint, reduce_on_plateau, tensorboard]
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(
monitor = 'loss',
min_delta = 0.01,
patience = 5,
mode = 'min',
verbose = 1
)
checkpoint = CustomModelCheckpoint(
model_to_save = model_to_save,
filepath = saved_weights_name,# + '{epoch:02d}.h5',
monitor = 'loss',
verbose = 1,
save_best_only = True,
mode = 'min',
period = 1
)
reduce_on_plateau = ReduceLROnPlateau(
monitor = 'loss',
factor = 0.1,
patience = 2,
verbose = 1,
mode = 'min',
epsilon = 0.01,
cooldown = 0,
min_lr = 0
)
tensorboard = CustomTensorBoard(
log_dir = tensorboard_logs,
write_graph = True,
write_images = True,
)
return [early_stop, checkpoint, reduce_on_plateau, tensorboard]
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.35, 0.35
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
# the main loop
for image_path in image_paths:
image = cv2.imread(image_path)
print(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
filename = output_path + image_path.split('/')[-1] + ".pkl"
outfile = open(filename, 'wb')
pickle.dump(boxes, outfile)
outfile.close()
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1], np.uint8(image))
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(monitor='loss',
min_delta=0.01,
patience=7,
mode='min',
verbose=1)
checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=saved_weights_name, # + '{epoch:02d}.h5',
monitor='loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
reduce_on_plateau = ReduceLROnPlateau(monitor='loss',
factor=0.1,
patience=2,
verbose=1,
mode='min',
epsilon=0.01,
cooldown=0,
min_lr=0)
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
# Learning rate reduce
reduce_on_plateau = ReduceLROnPlateau(
monitor="loss", # val_accuracy, loss
factor=0.1,
patience=7,
verbose=1,
mode="min",
cooldown=0,
min_lr=0,
min_delta=0.01,
)
# Tensorboard log callback
tb_cb = tf.keras.callbacks.TensorBoard(log_dir=tensorboard_logs,
histogram_freq=0,
write_graph=True,
profile_batch=0)
# Model checkpoint callback
checkpoint_cb = ModelCheckpoint(saved_weights_name,
monitor="loss",
verbose=1,
save_best_only=True,
save_weights_only=True,
mode="min",
period=1)
return [reduce_on_plateau, checkpoint_cb]
def train_data(self, gen, total=10000):
weights_path = os.path.join(self.tuner_dir, 'final-weights.pkl')
if os.path.isfile(weights_path):
with open(weights_path, 'rb') as fl:
res = pickle.load(fl)
logger.log_info('loaded weights from', weights_path,
highlight=4)
return res
logger.start()
iodata_path = os.path.join(self.tuner_dir, 'iodata.pkl')
if os.path.isfile(iodata_path):
with open(iodata_path, 'rb') as fl:
X, Y = pickle.load(fl)
logger.log_info('loaded io data from', iodata_path,
highlight=4)
else:
logger.log_info('constructing iodata', highlight=4)
X = []
Y = []
sz = 0
last = None
gen = take_first_n(gen, total)
# chunksize = (total + 4 * NUM_THREADS - 1) // (4 * NUM_THREADS)
if NUM_THREADS > 1:
with closing(multiprocessing.Pool(NUM_THREADS, maxtasksperchild=4)) as pool:
for i, (x, y) in tqdm(enumerate(pool.imap(
self.get_data, gen)), total=total):
X.append(x)
Y.append(y)
sz += len(x)
if last is None or sz > last + 50000:
logger.log_debug('%d/%d' % (i + 1, total), "with",
sz, "examples so far..")
last = sz
pool.close()
pool.join()
else:
for i, (x, y) in tqdm(enumerate(map(
self.get_data, gen)), total=total):
X.append(x)
Y.append(y)
sz += len(x)
if last is None or sz > last + 50000:
logger.log_debug('%d/%d' % (i + 1, total), "with",
sz, "examples so far..")
last = sz
X = np.concatenate(X, axis=0)
Y = np.concatenate(Y, axis=0)
makedirs(iodata_path)
with open(iodata_path, 'wb') as fl:
pickle.dump((X, Y), fl)
logger.log_info('dumping io data into', iodata_path,
highlight=4)
assert self.fix_delimiters_only, "The possibilities are not handled for non delimiters"
return self.fit(X, Y)
def predict(self, pic):
ret = []
with open(self.config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(self.output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.7, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
image_paths = []
if os.path.isdir(self.input_path):
for inp_file in os.listdir(self.input_path):
image_paths += [self.input_path + inp_file]
else:
image_paths += [self.input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
# the main loop
# for image_path in image_paths:
image = cv2.imread(pic)
print(pic)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
for box in boxes:
for i in range(len(config['model']['labels'])):
if box.classes[i] > obj_thresh:
ret.append([config['model']['labels'][i], box.classes[i]])
print(config['model']['labels'][i], " ", box.classes[i])
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
print("file save to {}".format(self.output_path + pic.split('/')[-1]))
cv2.imwrite(self.output_path + pic.split('/')[-1], np.uint8(image))
return ret
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save, valid_generator, config):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(
monitor = 'loss',
min_delta = 0.01,
patience = 5,
mode = 'min',
verbose = 1
)
checkpoint = CustomModelCheckpoint(
model_to_save = model_to_save,
filepath = saved_weights_name + '-epoch_{epoch:02d}-val_loss_{val_loss:.04f}.h5',
monitor = 'val_loss',
verbose = 1,
save_best_only = True,
mode = 'min',
period = 1
)
my_checkpoint = My_Checkpoint(
model_to_save=model_to_save,
valid_generator = valid_generator,
config = config
)
only_one_checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=saved_weights_name + '.h5',
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
period=1
)
reduce_on_plateau = ReduceLROnPlateau(
monitor = 'loss',
factor = 0.1,
patience = 2,
verbose = 1,
mode = 'min',
epsilon = 0.01,
cooldown = 0,
min_lr = 0
)
# tensorboard = CustomTensorBoard(
# log_dir = tensorboard_logs,
# write_graph = True,
# write_images = True,
# )
return [early_stop, checkpoint, only_one_checkpoint, reduce_on_plateau, my_checkpoint]
def _main_(config_path, input_path, output_path):
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(settings.BASE_DIR +
config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
# do detection on an image or a set of images
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
# the main loop
for image_path in image_paths:
image = cv2.imread(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
# draw bounding boxes on the image using labels
image_new, label_result = draw_boxes(image, boxes,
config['model']['labels'],
obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1],
np.uint8(image_new))
return label_result, output_path + image_path.split('/')[-1]
def fit(self, X, Y):
weights_path = os.path.join(self.tuner_dir, 'final-weights.pkl')
logger.log_debug(X.shape, Y.shape, np.unique(Y, return_counts=True))
#logger.log_debug("\n", np.round(X.mean(axis=0), 5))
#logger.log_debug("\n", np.round(X.std(axis=0), 5))
msk_dels = (Y[:, 0] == 0) | (Y[:, 0] == 1)
msk_adds = (Y[:, 0] == 2) | (Y[:, 0] == 3)
X_adds = X[msk_adds]
X_adds = X_adds[:, np.array([3, 4, 5]), ...]
Y_adds = Y[msk_adds]
Y_adds[Y_adds == 2] = 1
Y_adds[Y_adds == 3] = 0
combiner_adds = np.array(
[[1, 0],
[1, 0],
[0, 1]])
X_dels = X[msk_dels]
X_dels = X_dels[:, np.array([0, 1, 2]), ...]
Y_dels = Y[msk_dels]
Y_dels[Y_dels == 0] = 1
Y_dels[Y_dels == 1] = 0
combiner_dels = np.array(
[[1, 0],
[1, 0],
[0, 1]])
kernel_dels, bias_dels = self.create_network(X_dels, Y_dels, combiner_dels)
kernel_adds, bias_adds = self.create_network(X_adds, Y_adds, combiner_adds)
weights = np.ones(X.shape[1:])
bias = np.zeros(X.shape[1:])
weights[np.array([0, 1, 2])] = kernel_dels
bias[np.array([0, 1, 2])] = bias_dels
weights[np.array([3, 4, 5])] = kernel_adds
bias[np.array([3, 4, 5])] = bias_adds
logger.log_debug("\n", np.round(weights, 3), "\n", np.round(bias, 3))
#weights, bias = sparse.get_weights()
makedirs(weights_path)
with open(weights_path, 'wb') as fl:
pickle.dump((weights, bias), fl)
logger.log_info("logged weights into", weights_path, highlight=4)
logger.log_full_report_into_file(weights_path)
return weights, bias
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
anchors = []
with open('anchors.json') as anchors_str:
anchors = json.load(anchors_str)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.4, 0.45
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
infer_model = load_model(config['train']['saved_weights_name'])
# the main loop
for image_path in image_paths:
image = cv2.imread(image_path)
print(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w, anchors,
obj_thresh, nms_thresh)[0]
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1], np.uint8(image))
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(monitor='loss',
min_delta=0.01,
patience=5,
mode='min',
verbose=1)
checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=
'log_voc/ep{epoch:03d}-loss{loss:.3f}.h5', #saved_weights_name,# + '{epoch:02d}.h5',
monitor='loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
checkpoint2 = CustomModelCheckpoint2(
model_to_save=model_to_save,
filepath=saved_weights_name, # + '{epoch:02d}.h5',
monitor='loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
checkpoint3 = ModelCheckpoint('log_voc/ep{epoch:03d}-loss{loss:.3f}.h5',
monitor='loss',
save_best_only=True,
period=4)
#checkpoint3 = ModelCheckpoint('log_voc/ep{epoch:03d}-loss{loss:.3f}.h5', monitor='loss', save_weights_only=True, save_best_only=True, period=4)
#checkpoint2 not working yet for some reason... 1/15/2020
reduce_on_plateau = ReduceLROnPlateau(monitor='loss',
factor=0.1,
patience=2,
verbose=1,
mode='min',
epsilon=0.01,
cooldown=0,
min_lr=0)
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
return [checkpoint, checkpoint2, tensorboard]
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save):
makedirs(tensorboard_logs)
logging.debug("Adding callback (Early Stopping)")
early_stop = EarlyStopping(monitor='val_loss',
min_delta=0.1,
patience=8,
mode='min',
verbose=1,
restore_best_weights=True)
logging.debug("Adding callback (CustomModelCheckpoint)")
checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=saved_weights_name, # + '{epoch:02d}.h5',
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
logging.debug("Adding callback (ReduceLROnPlateau)")
reduce_on_plateau = ReduceLROnPlateau(monitor='val_loss',
factor=0.2,
patience=5,
verbose=1,
mode='min',
epsilon=0.01,
cooldown=3,
min_lr=0.0)
logging.debug("Adding callback (CustomTensorBoard)")
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
logging.debug("Adding callback (TerminateOnNaN)")
nan = tf.keras.callbacks.TerminateOnNaN()
logging.debug("Adding callback (LRTensorBoard)")
lr = LRTensorBoard(log_dir=tensorboard_logs)
return [early_stop, checkpoint, reduce_on_plateau, tensorboard, nan, lr]
def create_callbacks(model_to_save, config, valid_generator):
saved_weights_name = config['train']['saved_weights_name']
tensorboard_logs = config['train']['tensorboard']["log_dir"]
makedirs(tensorboard_logs)
early_stop = EarlyStopping(monitor='loss',
min_delta=0.01,
patience=5,
mode='min',
verbose=1)
checkpoint = CustomModelCheckpoint(
model_to_save=model_to_save,
filepath=saved_weights_name, # + '{epoch:02d}.h5',
monitor='loss',
verbose=1,
save_best_only=True,
mode='min',
period=1)
reduce_on_plateau = ReduceLROnPlateau(monitor='loss',
factor=0.1,
patience=2,
verbose=1,
mode='min',
epsilon=0.01,
cooldown=0,
min_lr=0)
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
tensorboardImage = TensorBoardImage(
tag=config["train"]["tensorboard"]["tag"],
labels=config["model"]["labels"],
infer_model=model_to_save,
valid_generator=valid_generator,
log_dir=tensorboard_logs,
anchors=config["model"]["anchors"],
img_count=config["train"]["tensorboard"]["img_count"])
return [
early_stop, checkpoint, reduce_on_plateau, tensorboard,
tensorboardImage
]
def rankGame(inputPath, draw_output=False):
with open(configPath) as config_buffer:
config = json.load(config_buffer)
makedirs(outputPath)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(
os.path.join(curDir, config['train']['saved_weights_name']))
image = cv2.imread(inputPath)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
# Sort boxes by x position sort in ascending player order
sortedBoxes = sorted(boxes, key=lambda box: box.xmin)
# draw bounding boxes on the image using labels
if draw_output:
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
# cv2.imwrite(outputPath.split('/')[-1], np.uint8(image))
cv2.imshow('Annotated Image', image)
cv2.waitKey(0)
# Assign ranks to the boxes based on their labels
rankedBoxes = rankBoxes(image, boxes, config['model']['labels'],
obj_thresh)
sortedRanks = [box.playerRank for box in rankedBoxes]
return sortedRanks # Returns a list of the integer ranks in order of ascending player number
def create_callbacks(saved_weights_name, tensorboard_logs, model_to_save,
textedit, cure, loss):
makedirs(tensorboard_logs)
early_stop = EarlyStopping(monitor='loss',
min_delta=0.01,
patience=5,
mode='min',
verbose=1)
checkpoint = CustomModelCheckpoint(
textedit,
cure,
loss,
model_to_save=model_to_save,
filepath='YOLO3_{epoch:02d}.h5', #saved_weights_name +,
monitor='loss',
verbose=1,
#save_best_only = True,
save_weights_only=False,
mode='min',
period=5)
reduce_on_plateau = ReduceLROnPlateau(monitor='loss',
factor=0.5,
patience=8,
verbose=1,
mode='min',
epsilon=0.01,
cooldown=0,
min_lr=0)
tensorboard = CustomTensorBoard(
log_dir=tensorboard_logs,
write_graph=True,
write_images=True,
)
return [ #early_stop,
checkpoint,
#reduce_on_plateau,
tensorboard
]
def main(args):
utils.makedirs(args.save_dir)
if args.print_to_log:
sys.stdout = open(f'{args.save_dir}/log.txt', 'w')
t.manual_seed(seed)
if t.cuda.is_available():
t.cuda.manual_seed_all(seed)
device = t.device('cuda' if t.cuda.is_available() else 'cpu')
model_cls = F if args.uncond else CCF
f = model_cls(args.depth, args.width, args.norm)
print(f"loading model from {args.load_path}")
# load em up
ckpt_dict = t.load(args.load_path)
f.load_state_dict(ckpt_dict["model_state_dict"])
replay_buffer = ckpt_dict["replay_buffer"]
f = f.to(device)
if args.eval == "OOD":
OODAUC(f, args, device)
if args.eval == "test_clf":
test_clf(f, args, device)
if args.eval == "cond_samples":
cond_samples(f, replay_buffer, args, device, args.fresh_samples)
if args.eval == "uncond_samples":
uncond_samples(f, args, device)
if args.eval == "logp_hist":
logp_hist(f, args, device)
def main(args, config):
if args.mode == "preprocess":
from preprocess import preprocess
try:
utils.makedirs("data/", raise_error=True)
except Exception:
print(
f"Warning : data folder already exists. Some data may get overwritten"
)
print("Running preprocessing")
preprocess(args, config)
elif args.mode == "train":
try:
utils.makedirs(args.output_folder, raise_error=True)
except Exception:
print(
f"Warning : {args.output_folder} already exists. Some data may get overwritten"
)
shutil.rmtree(f"{args.output_folder}/src", ignore_errors=True)
shutil.copytree("src/", f"{args.output_folder}/src")
shutil.copy(args.config_file, args.output_folder)
train(args, config)
elif args.mode == "test":
if args.model_file is None or args.test_file is None:
print("Error: Provide model_file and test_file file")
sys.exit(1)
try:
utils.makedirs(args.output_folder, raise_error=True)
except Exception:
print(
f"Warning : {args.output_folder} already exists. Some data may get overwritten"
)
test(args, config)
def save_model(self, save_path=None):
if save_path is None:
save_path = self.model_save_path
makedirs(save_path)
self.model.save(save_path)
logger.log_info(save_path, 'saved..')
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
if 'webcam' in input_path: # do detection on the first webcam
video_reader = cv2.VideoCapture(0)
# the main loop
batch_size = 1
images = []
while True:
ret_val, image = video_reader.read()
if ret_val == True: images += [image]
if (len(images)==batch_size) or (ret_val==False and len(images)>0):
batch_boxes = get_yolo_boxes(infer_model, images, net_h, net_w, config['model']['anchors'], obj_thresh, nms_thresh)
for i in range(len(images)):
draw_boxes(images[i], batch_boxes[i], config['model']['labels'], obj_thresh)
cv2.imshow('video with bboxes', images[i])
images = []
if cv2.waitKey(1) == 27:
break # esc to quit
cv2.destroyAllWindows()
elif input_path[-4:] == '.mp4': # do detection on a video
video_out = output_path + input_path.split('/')[-1]
video_reader = cv2.VideoCapture(input_path)
nb_frames = int(video_reader.get(cv2.CAP_PROP_FRAME_COUNT))
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
video_writer = cv2.VideoWriter(video_out,
cv2.VideoWriter_fourcc(*'MPEG'),
50.0,
(frame_w, frame_h))
# the main loop
batch_size = 1
images = []
start_point = 0 #%
show_window = False
for i in tqdm(range(nb_frames)):
_, image = video_reader.read()
if (float(i+1)/nb_frames) > start_point/100.:
images += [image]
if (i%batch_size == 0) or (i == (nb_frames-1) and len(images) > 0):
# predict the bounding boxes
batch_boxes = get_yolo_boxes(infer_model, images, net_h, net_w, config['model']['anchors'], obj_thresh, nms_thresh)
for i in range(len(images)):
# draw bounding boxes on the image using labels
draw_boxes(images[i], batch_boxes[i], config['model']['labels'], obj_thresh)
# show the video with detection bounding boxes
if show_window: cv2.imshow('video with bboxes', images[i])
# write result to the output video
video_writer.write(images[i])
images = []
if show_window and cv2.waitKey(1) == 27: break # esc to quit
if show_window: cv2.destroyAllWindows()
video_reader.release()
video_writer.release()
else: # do detection on an image or a set of images
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [inp_file for inp_file in image_paths if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])]
# the main loop
for image_path in image_paths:
image = cv2.imread(image_path)
print(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w, config['model']['anchors'], obj_thresh, nms_thresh)[0]
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1], np.uint8(image))
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
if 'webcam' in input_path: # do detection on the first webcam
video_reader = cv2.VideoCapture(0)
# the main loop
batch_size = 1
images = []
while True:
ret_val, image = video_reader.read()
if ret_val == True: images += [image]
if (len(images) == batch_size) or (ret_val == False
and len(images) > 0):
batch_boxes = get_yolo_boxes(infer_model, images, net_h, net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for i in range(len(images)):
draw_boxes(images[i], batch_boxes[i],
config['model']['labels'], obj_thresh)
cv2.imshow('video with bboxes', images[i])
images = []
if cv2.waitKey(1) == 27:
break # esc to quit
cv2.destroyAllWindows()
elif input_path[-4:] == '.mp4' or input_path[
-4:] == '.AVI': # do detection on a video
video_out = output_path + input_path.split('/')[-1].replace(
'.AVI', '.mp4')
video_reader = cv2.VideoCapture(input_path)
nb_frames = int(video_reader.get(cv2.CAP_PROP_FRAME_COUNT))
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
video_writer = cv2.VideoWriter(
video_out,
cv2.VideoWriter_fourcc(*'MP4V'), #(*'MPEG'),
50.0,
(frame_w, frame_h))
# the main loop
batch_size = 1
images = []
start_point = 0 #%
show_window = False
for i in tqdm(range(nb_frames)):
_, image = video_reader.read()
if (float(i + 1) / nb_frames) > start_point / 100.:
images += [image]
if (i % batch_size == 0) or (i == (nb_frames - 1)
and len(images) > 0):
# predict the bounding boxes
batch_boxes = get_yolo_boxes(infer_model, images, net_h,
net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for i in range(len(images)):
bbox0 = [batch_boxes[i][0]] if len(
batch_boxes[i]) else []
# draw bounding boxes on the image using labels
#draw_boxes(images[i], batch_boxes[i], config['model']['labels'], obj_thresh)
draw_boxes(images[i], bbox0, config['model']['labels'],
obj_thresh) # take only 1st bbox
# show the video with detection bounding boxes
if show_window:
cv2.imshow('video with bboxes', images[i])
# write result to the output video
video_writer.write(images[i])
images = []
if show_window and cv2.waitKey(1) == 27: break # esc to quit
if show_window: cv2.destroyAllWindows()
video_reader.release()
video_writer.release()
else: # do detection on an image or a set of images
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG', '.JPG'])
]
# the main loop
for image_path in image_paths:
#image_path = '/dataset/RZSS_images/1_animal_empty_r/animal/6.JPG'
image = cv2.imread(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
if len(boxes) > 0:
pboxes = np.array(
[[box.xmin, box.ymin, box.xmax, box.ymax,
box.get_score()] for box in boxes])
print(pboxes)
#print(boxes[0].xmin, boxes[0].ymin, boxes[0].xmax, boxes[0].ymax, boxes[0].c, boxes[0].classes );
#for k in range(len(boxes)): print(boxes[k].__dict__); import sys; sys.exit(0)
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
#import sys; sys.exit(0)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1],
np.uint8(image))
print('OUTPUT SAVED AS ' + output_path + image_path.split('/')[-1])
import json
import cv2
from utils.utils import get_yolo_boxes, makedirs
from utils.bbox import draw_boxes
from keras.models import load_model
from tqdm import tqdm
import numpy as np
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
downsample = 32 # ratio between network input's size and network output's size, 32 for YOLOv3
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
# this only works for squared images
if config['model']['min_input_size'] == config['model']['max_input_size']:
net_w = config['model']['min_input_size'] // downsample * downsample
net_h = config['model']['min_input_size'] // downsample * downsample
obj_thresh = config['train']['ignore_thresh']
nms_thresh = 0.45
if config['valid']['duplicate_thresh']:
nms_thresh = config['valid']['duplicate_thresh']
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
if 'webcam' in input_path: # do detection on the first webcam
video_reader = cv2.VideoCapture(0)
# the main loop
batch_size = 1
images = []
while True:
ret_val, image = video_reader.read()
if ret_val == True: images += [image]
if (len(images) == batch_size) or (ret_val == False
and len(images) > 0):
batch_boxes = get_yolo_boxes(infer_model, images, net_h, net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for i in range(len(images)):
draw_boxes(images[i], batch_boxes[i],
config['model']['labels'], obj_thresh)
cv2.imshow('video with bboxes', images[i])
images = []
if cv2.waitKey(1) == 27:
break # esc to quit
cv2.destroyAllWindows()
elif input_path[-4:] == '.mp4': # do detection on a video
video_out = output_path + input_path.split('/')[-1]
video_reader = cv2.VideoCapture(input_path)
nb_frames = int(video_reader.get(cv2.CAP_PROP_FRAME_COUNT))
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
video_writer = cv2.VideoWriter(video_out,
cv2.VideoWriter_fourcc(*'MPEG'), 50.0,
(frame_w, frame_h))
# the main loop
batch_size = 1
images = []
start_point = 0 #%
show_window = False
for i in tqdm(range(nb_frames)):
_, image = video_reader.read()
if (float(i + 1) / nb_frames) > start_point / 100.:
images += [image]
if (i % batch_size == 0) or (i == (nb_frames - 1)
and len(images) > 0):
# predict the bounding boxes
batch_boxes = get_yolo_boxes(infer_model, images, net_h,
net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for i in range(len(images)):
# draw bounding boxes on the image using labels
draw_boxes(images[i], batch_boxes[i],
config['model']['labels'], obj_thresh)
# show the video with detection bounding boxes
if show_window:
cv2.imshow('video with bboxes', images[i])
# write result to the output video
video_writer.write(images[i])
images = []
if show_window and cv2.waitKey(1) == 27: break # esc to quit
if show_window: cv2.destroyAllWindows()
video_reader.release()
video_writer.release()
else: # do detection on an image or a set of images
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
# the main loop
for image_i, image_path in enumerate(image_paths):
image = cv2.imread(image_path)
if image_i > 0 and image_i % 50 == 0:
print(
'predicted {:4} images out of {:4} images in total'.format(
image_i, len(image_paths)))
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1],
np.uint8(image))
print('predicted all {:4} images'.format(len(image_paths)))
# training hyperparameters
parser.add_argument('--batch_size', type=int, default=16)
parser.add_argument('--epochs', type=int, default=300)
# others
parser.add_argument('--seed', type=int, default=1)
args = parser.parse_args()
args.work_dir = osp.dirname(osp.realpath(__file__))
args.data_fp = osp.join(args.work_dir, 'data', args.dataset)
args.out_dir = osp.join(args.work_dir, 'out', args.exp_name)
args.checkpoints_dir = osp.join(args.out_dir, 'checkpoints')
print(args)
utils.makedirs(args.out_dir)
utils.makedirs(args.checkpoints_dir)
writer = writer.Writer(args)
device = torch.device('cuda', args.device_idx)
torch.set_num_threads(args.n_threads)
# deterministic
torch.manual_seed(args.seed)
cudnn.benchmark = False
cudnn.deterministic = True
# load dataset
template_fp = osp.join('template', 'template.obj')
meshdata = MeshData(args.data_fp,
template_fp,
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = sorted([
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
])
# the main loop
with open(os.path.join(output_path, "ans.xml"), "w") as file:
file.write("<annotations>\n")
for image_path in image_paths:
image = cv2.imread(image_path)
print(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
filename = os.path.basename(image_path)
write_boxes(file, filename, image, boxes,
config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1],
np.uint8(image))
file.write("</annotations>\n")
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
predict_path = args.predict
if_show = args.show
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
makedirs(predict_path)
###############################
# Set some parameter
###############################
net_h, net_w = 512, 512 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45 #0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
# infer_model = load_model('backend')
print('load model')
###############################
# Predict bounding boxes
###############################
# do detection on an image or a set of images
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths if (inp_file[-4:] in ['.mhd'])
]
# the main loop
for image_path in image_paths:
print(image_path)
slice_i = 1
while slice_i < 1000:
slice_i += 1
print('slice:' + str(slice_i))
image = raw_reader(image_path, slice_i)
if image is None:
break
if if_show:
image_ini = image[..., 2]
max_pix = np.max(image_ini)
min_pix = np.min(image_ini)
# print(max_pix, min_pix)
image_ini, _ = img_windowing(image_ini, max_pix, min_pix)
# cv2.imshow('image_ini', image_ini)
# cv2.waitKey()
# image_ini = np.uint8(np.float64((image_ini + 1000) / 1800) * 255)
(imagename,
extension) = os.path.splitext(image_path.split('/')[-1])
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh, imagename)[0]
line = ''
textname = predict_path + imagename + '_' + str(slice_i) + '.txt'
if if_show and len(boxes) > 0:
# draw bounding boxes on the image using labels
# print('boxes:' + str(len(boxes)))
draw_boxes(
image_ini, boxes, sorted(config['model']['labels']),
obj_thresh,
output_path + imagename + '_' + str(slice_i) + '.jpg')
# write the image with bounding boxes to file
# cv2.imwrite(output_path + imagename + '_' + str(slice_i) + '.jpg', np.uint8(image_ini))
newline = get_box_info(line, boxes,
sorted(config['model']['labels']),
obj_thresh)
with open(textname, 'w') as f:
f.write(newline)
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Parse the annotations
###############################
train_ints, valid_ints, labels, max_box_per_image = create_training_instances(
config['train']['train_annot_folder'],
config['train']['train_image_folder'], config['train']['cache_name'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
print('\nTraining on: \t' + str(labels) + '\n')
###############################
# Create the generators
###############################
train_generator = BatchGenerator(
instances=train_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.3,
norm=normalize)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Create the model
###############################
if os.path.exists(config['train']['saved_weights_name']):
config['train']['warmup_epochs'] = 0
warmup_batches = config['train']['warmup_epochs'] * (
config['train']['train_times'] * len(train_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
train_model, infer_model = create_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=max_box_per_image,
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=warmup_batches,
ignore_thresh=config['train']['ignore_thresh'],
multi_gpu=multi_gpu,
saved_weights_name=config['train']['saved_weights_name'],
lr=config['train']['learning_rate'],
grid_scales=config['train']['grid_scales'],
obj_scale=config['train']['obj_scale'],
noobj_scale=config['train']['noobj_scale'],
xywh_scale=config['train']['xywh_scale'],
class_scale=config['train']['class_scale'],
)
###############################
# Kick off the training
###############################
callbacks = create_callbacks(config['train']['saved_weights_name'],
config['train']['tensorboard_dir'],
infer_model)
history = train_model.fit_generator(
generator=train_generator,
steps_per_epoch=len(train_generator) *
config['train']['train_times'], #(train_images/4)*train_times
epochs=config['train']['nb_epochs'] + config['train']['warmup_epochs'],
verbose=2 if config['train']['debug'] else 1,
callbacks=callbacks,
workers=4,
max_queue_size=8,
validation_data=valid_generator,
validation_steps=len(valid_generator) * config['valid']['valid_times'])
# Prepare folder to save in
time = datetime.now()
time_str = time.strftime("%Y%m%d-%H:%M")
cwd = os.getcwd()
save_path = os.path.join(cwd, time_str)
print('Data saved in: ' + save_path)
makedirs(save_path)
# Copy and save config file
save_config_path = os.path.join(save_path, 'config.json')
shutil.copy2(config_path, save_config_path)
# plotting train data and saving mat file with data
plot_matlab(history, config['data']['plot_png'],
config['data']['save_mat'], save_path)
# Save the loaded models to .TXT
save_template_path = os.path.join(save_path, 'train_model_params.txt')
with open(save_template_path, 'w') as fh:
# Pass the file handle in as a lambda function to make it callable
train_model.summary(print_fn=lambda x: fh.write(x + '\n'))
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
print('========== VALIDATION ==========')
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
dataformate = date.strftime("%Y-%m-%d %H:%M:%S") #上線版本
#day='20210331'
# In[root]
currentroot = 'D:/Camy/AI_project/CELL_ASM_SealWidth_SPC_N2/code' #絕對路徑，相對路徑os.getcwd()
os.chdir(currentroot)
#currentroot=os.getcwd()
down_img = currentroot + '/input/Downpic/'
input_img = currentroot + '/input/images/'
image_compare = currentroot + '/input/compare/'
output_img = currentroot + '/output/yoloimage/' #/image/
output_csv = currentroot + '/output/csv/' #/image/
#output_csv= currentroot+'/output/csv_data/'
#check_table=currentroot+'/CheckTable'
Models_path = currentroot + '/models/'
##Creat file
makedirs(down_img)
makedirs(input_img)
makedirs(image_compare)
makedirs(output_img)
makedirs(output_csv)
#makedirs(output_csv)
#makedirs(check_table)
# In[readConfig]
config = configparser.ConfigParser()
config.read('Config.ini', encoding='utf-8') #config檔案名稱
ModelName = list(config['Setting']['model_Name'].split(','))
ModelH5File = list(config['Setting']['model_h5Nmae'].split(','))
SheetModel_Modelname_dic = {}
for d in range(len(ModelName)):
SheetModel_Modelname_dic[ModelName[d]] = ModelH5File[d]
# In[Set Model some parameter]
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
train_data = pd.DataFrame(columns=[
'clip_id', 'x0', 'x1', 'x2', 'x3', 'x4', 'x5', 'x6', 'x7', 'x8', 'x9',
'x10', 'x11', 'x12', 'x13', 'x14', 'y0', 'y1', 'y2', 'y3', 'y4', 'y5',
'y6', 'y7', 'y8', 'y9', 'y10', 'y11', 'y12', 'y13', 'y14'
])
vid_files = [
f for f in os.listdir(input_path) if f[-4:] in [".avi", ".mp4"]
]
for num, vid_file in enumerate(vid_files):
print(
"#############\nProcessing on video No.{} out of {} videos: {}\n#############"
.format(num, len(vid_files), vid_file))
start = time.time()
video_out = output_path + '/traj_' + vid_file.split('/')[-1]
video_reader = cv2.VideoCapture(input_path + vid_file)
clip_id = int("".join(re.findall("[0-9]", vid_file)))
video_fps = int(video_reader.get(cv2.CAP_PROP_FPS))
nb_frames = int(video_reader.get(cv2.CAP_PROP_FRAME_COUNT))
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
# the main loop
batch_size = 1
images = []
start_point = 0 #%
show_window = False
balls_coords = []
balls_frms = []
labels = config['model']['labels']
for i in tqdm(range(nb_frames)):
_, image = video_reader.read()
# stack all frames
balls_frms.append(image)
if (float(i + 1) / nb_frames) > start_point / 100.:
images += [image]
if (i % batch_size == 0) or (i == (nb_frames - 1)
and len(images) > 0):
# predict the bounding boxes
batch_boxes = get_yolo_boxes(infer_model, images, net_h,
net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for k in range(len(images)):
# stack all ball coords
coords = get_coord(images[k], batch_boxes[k], labels,
obj_thresh, i) # i: frame no
balls_coords = balls_coords + coords
# show the video after above operations
if show_window:
cv2.imshow('video with bboxes', images[k])
images = []
if show_window and cv2.waitKey(1) == 27: break # esc to quit
# track trajectory
x_output, y_output = pitch_predict(balls_coords, balls_frms, video_out,
video_fps, frame_w, frame_h)
record = [clip_id] + x_output + y_output
if len(record) > 1:
train_data.loc[train_data.shape[0] - 1, :] = record
train_data.to_csv("output-0614-1.csv", index=False)
if show_window: cv2.destroyAllWindows()
video_reader.release()
print('\n>> Time spent on video No.{}, {}: {}seconds.'.format(
num, vid_file,
time.time() - start))
def _main_(args):
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
json_file = open('model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
loaded_model.load_weights("model.h5")
label_map = np.load('label_map.npy').item()
###############################
# Predict bounding boxes
###############################
if 'webcam' in input_path: # do detection on the first webcam
video_reader = cv2.VideoCapture(0)
# the main loop
batch_size = 1
images = []
while True:
ret_val, image = video_reader.read()
if ret_val == True:
images += [image]
if (len(images) == batch_size) or (ret_val == False
and len(images) > 0):
batch_boxes = get_yolo_boxes(infer_model, images, net_h, net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for i in range(len(images)):
draw_boxes(images[i], batch_boxes[i],
config['model']['labels'], obj_thresh)
cv2.imshow('video with bboxes', images[i])
images = []
if cv2.waitKey(1) == 27:
break # esc to quit
cv2.destroyAllWindows()
elif input_path[-4:] == '.mp4': # do detection on a video
video_out = output_path + input_path.split('/')[-1]
video_reader = cv2.VideoCapture(input_path)
nb_frames = int(video_reader.get(cv2.CAP_PROP_FRAME_COUNT))
frame_h = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
frame_w = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
video_writer = cv2.VideoWriter(video_out,
cv2.VideoWriter_fourcc(*'MPEG'), 50.0,
(frame_w, frame_h))
# the main loop
batch_size = 1
images = []
start_point = 0 # %
show_window = False
for i in tqdm(range(nb_frames)):
_, image = video_reader.read()
if (float(i + 1) / nb_frames) > start_point / 100.:
images += [image]
if (i % batch_size == 0) or (i == (nb_frames - 1)
and len(images) > 0):
# predict the bounding boxes
batch_boxes = get_yolo_boxes(infer_model, images, net_h,
net_w,
config['model']['anchors'],
obj_thresh, nms_thresh)
for i in range(len(images)):
# draw bounding boxes on the image using labels
draw_boxes(images[i], batch_boxes[i], loaded_model,
label_map, config['model']['labels'],
obj_thresh)
# show the video with detection bounding boxes
if show_window:
cv2.imshow('video with bboxes', images[i])
# write result to the output video
video_writer.write(images[i])
images = []
if show_window and cv2.waitKey(1) == 27:
break # esc to quit
if show_window:
cv2.destroyAllWindows()
video_reader.release()
video_writer.release()
else: # do detection on an image or a set of images
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
print(input_path + inp_file)
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
# the main loop
for image_path in tqdm(image_paths):
# print(image_path)
image = cv2.imread(image_path)
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
# draw bounding boxes on the image using labels
draw_boxes(image, boxes, loaded_model, label_map,
sorted(config['model']['labels']), obj_thresh)
# write the image with bounding boxes to file
cv2.imwrite(output_path + image_path.split('/')[-1],
np.uint8(image))
def work(textedit, pic_label, input, model, output):
input = input + "/"
output = output + "/"
argparser = argparse.ArgumentParser(
description='Predict with a trained yolo model')
argparser.add_argument('-c',
'--conf',
help='path to configuration file',
default='config1.json')
argparser.add_argument(
'-i',
'--input',
help='path to an image, a directory of images, a video, or webcam',
default=input)
argparser.add_argument('-o',
'--output',
default=output,
help='path to output directory')
args = argparser.parse_args()
#_main_(args)
config_path = args.conf
input_path = args.input
output_path = args.output
with open(config_path) as config_buffer:
config = json.load(config_buffer)
makedirs(output_path)
###############################
# Set some parameter
###############################
net_h, net_w = 416, 416 # a multiple of 32, the smaller the faster
obj_thresh, nms_thresh = 0.5, 0.45
###############################
# Load the model
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
#config['train']['saved_weights_name']
keras.backend.clear_session()
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Predict bounding boxes
###############################
image_paths = []
if os.path.isdir(input_path):
for inp_file in os.listdir(input_path):
image_paths += [input_path + inp_file]
else:
image_paths += [input_path]
image_paths = [
inp_file for inp_file in image_paths
if (inp_file[-4:] in ['.jpg', '.png', 'JPEG'])
]
# the main loop
strideN = 208 ##步长
for image_path in image_paths:
imageO = cv2.imread(image_path)
print(image_path)
image = QtGui.QPixmap(image_path)
pic_label.setPixmap(image)
pic_label.setScaledContents(True)
(height, width, _) = imageO.shape
mH = int((height - strideN) / strideN)
mW = int((width - strideN) / strideN)
####对图像进行分割处理，网格搜索
object_key = []
object_pro = []
object_x1 = []
object_y1 = []
object_x2 = []
object_y2 = []
for m in range(mH):
for n in range(mW):
print(m * mW + n)
cursor = textedit.textCursor()
cursor.movePosition(QtGui.QTextCursor.End)
cursor.insertText("Detecting: " + str(m * mW + n))
cursor.insertText("\r\n")
# textedit.append('Elapsed time = {}'.format(time.time() - st))
textedit.setTextCursor(cursor)
textedit.ensureCursorVisible()
flag = False
'''
cursor = textedit.textCursor()
cursor.movePosition(QtGui.QTextCursor.End)
cursor.insertText(str(m * mW + n))
cursor.insertText("\r\n")
'''
imgCopy = imageO.copy()
image = imgCopy[strideN * m:strideN * (m + 2),
strideN * n:strideN * (n + 2)] ##height,width
# predict the bounding boxes
boxes = get_yolo_boxes(infer_model, [image], net_h, net_w,
config['model']['anchors'], obj_thresh,
nms_thresh)[0]
# draw bounding boxes on the image using labels
#draw_boxes(image, boxes, config['model']['labels'], obj_thresh)
# write the image with bounding boxes to file
#aaa=image_path.split('/')
#aab=aaa[-1].split('.')
#cv2.imwrite(output_path + aab[0] + '_' + str(m*mW+n+1) + '.' + aab[1], np.uint8(image))
####存储所有的检测框
labels = config['model']['labels']
quiet = True
key = []
prob = []
for box in boxes:
label_str = ''
label = -1
for i in range(len(labels)):
if box.classes[i] > obj_thresh:
if label_str != '': label_str += ', '
label_str += (
labels[i] + ' ' +
str(round(box.get_score() * 100, 2)) + '%'
) ###概率值
key = labels[i]
prob = box.get_score()
label = i
if not quiet: print(label_str)
color = [0, 0, 255]
if key == "missile": color = [0, 0, 255]
if key == "oiltank": color = [0, 159, 255]
if key == "plane": color = [0, 255, 0]
if key == "warship": color = [255, 0, 0]
if label >= 0:
flag = True
text_size = cv2.getTextSize(label_str,
cv2.FONT_HERSHEY_SIMPLEX,
1.1e-3 * image.shape[0], 5)
width, height = text_size[0][0], text_size[0][1]
region = np.array(
[[box.xmin - 3, box.ymin],
[box.xmin - 3, box.ymin - height - 26],
[box.xmin + width + 13, box.ymin - height - 26],
[box.xmin + width + 13, box.ymin]],
dtype='int32')
cv2.rectangle(image, (box.xmin, box.ymin),
(box.xmax, box.ymax),
color,
thickness=2)
aaa = image_path.split('/')
aab = aaa[-1].split('.')
cv2.imwrite(
output_path + aab[0] + '_' + str(m * mW + n + 1) +
'.' + aab[1], np.uint8(image))
print(output_path + aab[0] + '_' +
str(m * mW + n + 1) + '.' + aab[1])
#image = QtGui.QPixmap(output_path + aab[0] + '_' + str(m*mW+n+1) + '.' + aab[1])
#pic_label.setPixmap(image)
#pic_label.setScaledContents(True)
object_real_x1 = box.xmin + strideN * n
object_real_y1 = box.ymin + strideN * m
object_real_x2 = box.xmax + strideN * n
object_real_y2 = box.ymax + strideN * m
object_key.append(key)
object_pro.append(prob)
object_x1.append(object_real_x1)
object_y1.append(object_real_y1)
object_x2.append(object_real_x2)
object_y2.append(object_real_y2)
if flag:
flag = False
aaa = image_path.split('/')
aab = aaa[-1].split('.')
image = QtGui.QPixmap(output_path + aab[0] + '_' +
str(m * mW + n + 1) + '.' + aab[1])
pic_label.setPixmap(image)
pic_label.setScaledContents(True)
##非极大值抑制
imgCopy2 = imageO.copy()
object_name = ["missile", "oiltank", "plane", "warship"]
for object_class in range(len(object_name)):
x1 = []
y1 = []
x2 = []
y2 = []
prob = []
for numR in range(len(object_key)):
if object_key[numR] == object_name[object_class]:
x1.append(object_x1[numR])
y1.append(object_y1[numR])
x2.append(object_x2[numR])
y2.append(object_y2[numR])
prob.append(object_pro[numR])
if len(x1) > 0:
x1 = np.array(x1)
y1 = np.array(y1)
x2 = np.array(x2)
y2 = np.array(y2)
prob = np.array(prob)
x1, y1, x2, y2, probs = non_max_suppression(x1,
y1,
x2,
y2,
prob,
overlap_thresh=0.5,
max_boxes=30)
for numLR in range(len(x1)):
real_x1 = x1[numLR]
real_y1 = y1[numLR]
real_x2 = x2[numLR]
real_y2 = y2[numLR]
color = [0, 0, 255]
if object_name[object_class] == "missile":
color = [0, 0, 255]
if object_name[object_class] == "oiltank":
color = [0, 159, 255]
if object_name[object_class] == "plane":
color = [0, 255, 0]
if object_name[object_class] == "warship":
color = [255, 0, 0]
cv2.rectangle(imgCopy2, (real_x1, real_y1),
(real_x2, real_y2), color, 2)
cv2.imwrite(output_path + image_path.split('/')[-1], imgCopy2)
image = QtGui.QPixmap(output_path + image_path.split('/')[-1])
pic_label.setPixmap(image)
pic_label.setScaledContents(True)