def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--data_dir', default="data", type=str)
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
if args.weights is not None:
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet()
pascal = pascal_voc('train', True)
pascal_test = pascal_voc('test', True)
solver = Solver(yolo, pascal, pascal_test)
print('Start training ...')
solver.train()
print('Done training.')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="D:\\reference\\5-dataset\\YOLO_small.ckpt", type=str) # 预训练的数据集
parser.add_argument('--data_dir', default="D:\\reference\\5-dataset", type=str) # voc数据集文件夹pascal_voc所在路径
parser.add_argument('--threshold', default=0.2, type=float) # 格子有目标的置信度阈值
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='0', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU # 获取系统环境变量
yolo = YOLONet()
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
def main(): #自定义参数
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt",
type=str) #定义权重文件
parser.add_argument('--data_dir', default="data", type=str) #定义数据文件夹
parser.add_argument('--threshold', default=0.2, type=float) #阈值
parser.add_argument('--iou_threshold', default=0.5, type=float) #IOU阈值
parser.add_argument('--gpu', default='', type=str) #是否用gpu训练
args = parser.parse_args()
if args.gpu is not None: #是否用gpu训练
cfg.GPU = args.gpu
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet() #Yolo网络
pascal = pascal_voc('train') #获得训练的数据， 包含了经过水平翻转后的训练实例
solver = Solver(yolo, pascal) #准备训练的环境，包括设置优化器，学习率等内容
print('Start training ...')
solver.train() #start training
print('done!!!')
def main():
# 创建一个解析器对象，并告诉它将会有些什么参数。当程序运行时，该解析器就可以用于处理命令行参数。
parser = argparse.ArgumentParser() # 定义参数
parser.add_argument('--weights', default="YOLO_small.ckpt",
type=str) # 权重文件名
parser.add_argument('--data_dir', default="data", type=str) # 数据集路径
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='', type=str)
# 定义了所有参数之后，你就可以给 parse_args() 传递一组参数字符串来解析命令行。默认情况下，参数是从 sys.argv[1:] 中获取
# parse_args() 的返回值是一个命名空间，包含传递给命令的参数。该对象将参数保存其属性
args = parser.parse_args()
# 判断是否是使用gpu
if args.gpu is not None:
cfg.GPU = args.gpu
# 设定数据集路径，以及检查点文件路径
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
# 设置环境变量
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet() # 创建YOLO网络对象
pascal = pascal_voc('train') # 数据集对象
solver = Solver(yolo, pascal) # 求解器对象
print('Start training ...')
solver.train() # 开始训练
print('Done training.')
def generate_GTbox_labels(rebuild=False, save=True):
voc_test = pascal_voc(phase='test', rebuild=rebuild)
result_dir = './result'
if not os.path.exists(result_dir):
os.mkdir(result_dir)
labels_info = voc_test.gt_labels
nums = len(labels_info)
# print(nums)
GT_box = []
GT_labels = []
if save:
for i, info in enumerate(labels_info):
# print(info['imname'])
lab = info['label'][..., 1:]
response = info['label'][..., 0]
index = np.nonzero(response)
# convert to (ymin,xmin,ymax,xmax)
xmin = lab[..., 0] - 0.5 * lab[..., 2]
ymin = lab[..., 1] - 0.5 * lab[..., 3]
xmax = lab[..., 0] + 0.5 * lab[..., 2]
ymax = lab[..., 1] + 0.5 * lab[..., 3]
bboxes = np.stack([ymin, xmin, ymax, xmax], axis=-1)[index]
GT_box.append(bboxes)
# print(GT_box)
classes = np.argmax(lab[..., 4:], axis=-1)[index]
GT_labels.append(classes)
#print(GT_box,GT_labels,sep='\n')
r = '\r>>>>>>Generate:{:4d}/{}'.format(i + 1, nums)
print(r, end='', flush=True)
np.save(os.path.join(result_dir, 'GT_box.npy'), GT_box)
np.save(os.path.join(result_dir, 'GT_labels.npy'), GT_labels)
return voc_test
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--data_dir', default="data", type=str)
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
yolo = YOLONet(True)
data = pascal_voc('train')
solver = Solver(yolo, data)
print('Start training ...')
solver.train()
print('Done training.')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--data_dir', default="data", type=str)
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet()
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
def main():
parser = argparse.ArgumentParser() # 创建一个解析器对象，并告诉它将会有些什么参数
# 当程序运行时，该解析器就可以用于处理命令行参数
'''定义参数'''
parser.add_argument('--weights', default="YOLO_small.ckpt",
type=str) # 权重文件名
parser.add_argument('--data_dir', default="data", type=str) # 数据集目录
parser.add_argument('--threshold', default=0.2, type=float) # 目标存在阈值
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='', type=str)
# 定义了所有参数之后，就可以给parse_args()传递一组参数字符串来解析命令行
# 默认情况下参数从sys.argv[1:]中获取
args = parser.parse_args() #将命令行中的参数保存相应的属性
if args.gpu is not None:
cfg.GPU = args.gpu # 判断是否使用GPU
if args.data_dir != cfg.DATA_PATH:
# 检查当前数据集路径，并进行更新
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU # 设置环境变量
yolo = YOLONet() #创建yolo
pascal = pascal_voc('train') #数据集
solver = Solver(yolo, pascal) # 求解器
print('Start training ...')
solver.train()
print('Done training.')
def main():
modelNet = cnnLstmNet()
pascal = pascal_voc('train')
solver = Solver(modelNet, pascal)
print('开始训练...')
solver.train()
print('训练结束....')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--gpu', default=None, type=int)
args = parser.parse_args()
cfg.WEIGHTS_FILE = os.path.join(cfg.WEIGHTS_DIR, args.weights)
if args.gpu is not None:
cfg.GPU = str(args.gpu)
yolo = YOLONet('train')
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
solver.train()
def main(args):
if args.gpu is not None:
cfg.GPU = args.gpu
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet()
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
def main():
tf.reset_default_graph()
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--data_dir', default="data", type=str)
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='0', type=str)
args = parser.parse_args()
os.environ['CUDA_VISIBLE_DEVICES'] = "0"
yolo = Yolo()
pascal = pascal_voc('test')
solver = Solver(yolo, pascal)
print('Start testing ...')
solver.train()
print('Done testing.')
def main():
#command line interface
parser = argparse.ArgumentParser()
parser.add_argument('-threshold', default=cfg.THRESHOLD, type=float)
parser.add_argument('-iou_threshold',
default=cfg.IOU_THRESHOLD,
type=float)
parser.add_argument('-gpu', default='0', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
trainer = Trainer(YOLO(), pascal_voc('train'))
print('Start training ...')
trainer.train()
print('Done training.')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--gpu', default='0', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
if args.weights is not None:
cfg.WEIGHTS = args.weights
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet()
pascal = pascal_voc()
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="YOLO_small.ckpt", type=str)
parser.add_argument('--data_dir', default="data", type=str)
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='', type=str)
args = parser.parse_args()
if args.gpu is not None:
cfg.GPU = args.gpu
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
yolo = YOLONet()
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
def main():
if not os.path.exists(cfg.OUTPUT_dir):
os.makedirs(cfg.OUTPUT_dir)
if not os.path.exists(cfg.TRAIN_process_save_txt_dir):
os.makedirs(cfg.TRAIN_process_save_txt_dir)
if not os.path.exists(cfg.WEIGHTS_output_dir):
os.makedirs(cfg.WEIGHTS_output_dir)
# if args.data_dir != cfg.DATA_PATH:
# update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet()
pascal = pascal_voc('train', rebuild=False)
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default=None, type=str)
parser.add_argument('--gpu', default=None, type=int)
args = parser.parse_args()
if args.weights is not None:
cfg.WEIGHTS_FILE = os.path.join(cfg.WEIGHTS_DIR, args.weights)
if args.gpu is not None:
cfg.GPU = str(args.gpu)
cfg.GPU = '1'
cfg.WEIGHTS_FILE = os.path.join(cfg.WEIGHTS_DIR, 'YOLO_small.ckpt')
# cfg.DISPLAY_ITER = 1
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU
yolo = YOLONet('train')
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
solver.train()
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', default="/home/jiahuei/Documents/Woh/pothole-detection/data/pascal_voc/output/2018_07_06_20_14/save.ckpt-6000", type=str)
parser.add_argument('--data_dir', default="data", type=str)
parser.add_argument('--threshold', default=0.2, type=float)
parser.add_argument('--iou_threshold', default=0.5, type=float)
parser.add_argument('--gpu', default='0', type=str)
args = parser.parse_args()
#if args.gpu is not None:
# cfg.GPU = args.gpu
if args.data_dir != cfg.DATA_PATH:
update_config_paths(args.data_dir, args.weights)
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
yolo = YOLONet()
pascal = pascal_voc('train')
solver = Solver(yolo, pascal)
print('Start training ...')
solver.train()
print('Done training.')
# -*- coding: utf-8 -*-
import tensorflow as tf
import tensorflow.keras as keras
import config as cfg
from YOLOv1_beta_1 import YOLOv1Net
from utils.pascal_voc import pascal_voc
# 需要首先下载数据集
pascal = pascal_voc('train')
def get_train_data_by_batch():
while 1:
for i in range(0, len(pascal.gt_labels), 64):
images, labels = pascal.get()
yield (images, labels)
def lr_scheduler(epoch):
lr = 1e-4
if(epoch <= 75):
lr = 1e-2
elif(75 < epoch and epoch <= 105):
lr = 1e-3
elif(105 < epoch and epoch <= 135):
lr = 1e-4
return lr
sess.run(
tf.assign(
self._variables_to_fix['vgg_16' + '/fc6/weights:0'],
tf.reshape(
fc6_conv, self._variables_to_fix[
'vgg_16' + '/fc6/weights:0'].get_shape())))
sess.run(
tf.assign(
self._variables_to_fix['vgg_16' + '/fc7/weights:0'],
tf.reshape(
fc7_conv, self._variables_to_fix[
'vgg_16' + '/fc7/weights:0'].get_shape())))
if __name__ == '__main__':
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU_ID
backbone = 'resnet'
net = network.Net()
rpn_loss_obj = RPN_loss(net.rois_output['rois_bbx'], net.all_anchors, net.gt_boxes, \
net.rois_output['rois_cls'], net.labels, net.anchor_obj)
predict_loss = Predict_loss(net._predictions["cls_score"], net._proposal_targets['labels'],\
net._predictions['bbox_pred'], net._proposal_targets['bbox_targets'],\
net._proposal_targets['bbox_inside_weights'], net._proposal_targets['bbox_outside_weights'])
train_data = pascl.pascal_voc(cfg.train_imdb_name, 'train', fliped=True)
val_data = pascl.pascal_voc(cfg.test_imdb_name, 'test', fliped=False)
solver = Solver(net, train_data, val_data, rpn_loss_obj, predict_loss)
print('start training')
solver.train_model(backbone)
if not R['dec_id'][jmax]:
tp[d] = 1.
R['dec_id'][jmax] = 1
else:
fp[d] = 1.
else:
fp[d] = 1.
fp = np.cumsum(fp)
tp = np.cumsum(tp)
rec = tp / float(num_cls_obj)
prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
ap.append(self.val_data.voc_ap(rec, prec))
print(np.mean(ap))
return sum(ap) / (self.net.num_classes - 1.0)
def get_var_list(self, global_variables, ckpt_variables):
variables_to_restore = []
for key in global_variables:
if key.name.split(':')[0] in ckpt_variables:
variables_to_restore.append(key)
return variables_to_restore
if __name__ == '__main__':
os.environ['CUDA_VISIBLE_DEVICES'] = cfg.GPU_ID
net = network.Net(is_training=False, backbone='resnet')
# net = network.Net(is_training=False, backbone='vgg16')
val_data = pascl.pascal_voc(cfg.test_imdb_name, 'test', fliped=False)
test = Val(net, val_data)
print('start training')
test.test_model()
