def main():
parser = argparse.ArgumentParser(description='Text Recognition Training')
parser.add_argument('exp', type=str)
parser.add_argument('--resume', type=str, help='Resume from checkpoint')
parser.add_argument('--image_path', type=str, help='image path')
parser.add_argument('--result_dir', type=str, default='./demo_results/', help='path to save results')
parser.add_argument('--data', type=str,
help='The name of dataloader which will be evaluated on.')
parser.add_argument('--image_short_side', type=int, default=736,
help='The threshold to replace it in the representers')
parser.add_argument('--thresh', type=float,
help='The threshold to replace it in the representers')
parser.add_argument('--box_thresh', type=float, default=0.6,
help='The threshold to replace it in the representers')
parser.add_argument('--visualize', action='store_true',
help='visualize maps in tensorboard')
parser.add_argument('--resize', action='store_true',
help='resize')
parser.add_argument('--polygon', action='store_true',
help='output polygons if true')
parser.add_argument('--eager', '--eager_show', action='store_true', dest='eager_show',
help='Show iamges eagerly')
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
experiment = Configurable.construct_class_from_config(experiment_args)
Demo(experiment, experiment_args, cmd=args).inference(args['image_path'], args['visualize'])
def main():
parser = argparse.ArgumentParser(description='Text Recognition Training')
parser.add_argument('--exp', type=str, default=exp)
parser.add_argument('--resume', type=str, help='Resume from checkpoint', default=ckpt_path)
parser.add_argument('--result_dir', type=str, default='./demo_results/', help='path to save results')
parser.add_argument('--data', type=str,
help='The name of dataloader which will be evaluated on.')
parser.add_argument('--image_short_side', type=int, default=img_short_side,
help='The threshold to replace it in the representers')
parser.add_argument('--thresh', type=float,
help='The threshold to replace it in the representers')
parser.add_argument('--box_thresh', type=float, default=detector_box_thres,
help='The threshold to replace it in the representers')
parser.add_argument('--resize', action='store_true', help='resize')
parser.add_argument('--visualize', default=visualize, help='visualize maps in tensorboard')
parser.add_argument('--polygon', help='output polygons if true', default=polygon)
parser.add_argument('--eager', '--eager_show', action='store_true', dest='eager_show',
help='Show iamges eagerly')
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
experiment = Configurable.construct_class_from_config(experiment_args)
detector = Demo(experiment, experiment_args, gpu=gpu_test, cmd=args)
begin = time.time()
detector.inference2(img_path, args['visualize'])
end = time.time()
print ( 'Exe time:',end-begin,'seconds')
def main():
parser = argparse.ArgumentParser(description='Text Recognition Training')
parser.add_argument('exp', type=str)
parser.add_argument('--name', type=str)
parser.add_argument('--batch_size', type=int, help='Batch size for training')
parser.add_argument('--resume', type=str, help='Resume from checkpoint')
parser.add_argument('--epochs', type=int, help='Number of training epochs')
parser.add_argument('--num_workers', type=int, help='Number of dataloader workers')
parser.add_argument('--start_iter', type=int,
help='Begin counting iterations starting from this value (should be used with resume)')
parser.add_argument('--start_epoch', type=int,
help='Begin counting epoch starting from this value (should be used with resume)')
parser.add_argument('--max_size', type=int, help='max length of label')
parser.add_argument('--lr', type=float, help='initial learning rate')
parser.add_argument('--optimizer', type=str, help='The optimizer want to use')
parser.add_argument('--thresh', type=float, help='The threshold to replace it in the representers')
parser.add_argument('--verbose', action='store_true', help='show verbose info')
parser.add_argument('--visualize', action='store_true', help='visualize maps in tensorboard')
parser.add_argument('--force_reload', action='store_true', dest='force_reload', help='Force reload data meta')
parser.add_argument('--no-force_reload', action='store_false', dest='force_reload', help='Force reload data meta')
parser.add_argument('--validate', action='store_true', dest='validate', help='Validate during training')
parser.add_argument('--no-validate', action='store_false', dest='validate', help='Validate during training')
parser.add_argument('--print-config-only', action='store_true', help='print config without actual training')
parser.add_argument('--debug', action='store_true', dest='debug',
help='Run with debug mode, which hacks dataset num_samples to toy number')
parser.add_argument('--no-debug', action='store_false', dest='debug', help='Run without debug mode')
parser.add_argument('--benchmark', action='store_true', dest='benchmark', help='Open cudnn benchmark mode')
parser.add_argument('--no-benchmark', action='store_false', dest='benchmark', help='Turn cudnn benchmark mode off')
parser.add_argument('-d', '--distributed', action='store_true', dest='distributed', help='Use distributed training')
parser.add_argument('--local_rank', dest='local_rank', default=0, type=int, help='Use distributed training')
parser.add_argument('-g', '--num_gpus', dest='num_gpus', default=4, type=int, help='The number of accessible gpus')
parser.set_defaults(debug=False)
parser.set_defaults(benchmark=True)
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
# Torch分布式训练，据称即使是单电脑多卡，也比data_parallel快，使用nccl后端（需要部署）
if args['distributed']:
torch.cuda.set_device(args['local_rank'])
torch.distributed.init_process_group(backend='nccl', init_method='env://')
conf = Config()
# 通过yaml文件，生成对应的配置参数，experiment里包含了name,class,structure,train,validation,logger,evaluation(=valid)信息
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
# 使用命令行参数覆盖一些默认参数，也就是命令行参数优先级高于默认参数
experiment_args.update(cmd=args)
# 通过参数表构筑实验，experiment里包含distributed,local_rank(分布式时有用),evaluation(训练时为空),logger,states,structure,train,validation
experiment = Configurable.construct_class_from_config(experiment_args)
if not args['print_config_only']:
torch.backends.cudnn.benchmark = args['benchmark']
# 构筑训练器
trainer = Trainer(experiment)
# 开始训练
trainer.train()
def main():
parser = argparse.ArgumentParser(description='Text Recognition Training')
parser.add_argument('--exp', type=str, default=config_file)
parser.add_argument('--name', type=str, default=training_dir)
parser.add_argument('--batch_size', type=int, help='Batch size for training')
parser.add_argument('--resume', type=str, default=resume_ckpt, help='Resume from checkpoint')
parser.add_argument('--epochs', type=int, help='Number of training epochs')
parser.add_argument('--num_workers', type=int, help='Number of dataloader workers')
parser.add_argument('--start_iter', type=int,
help='Begin counting iterations starting from this value (should be used with resume)')
parser.add_argument('--start_epoch', type=int,
help='Begin counting epoch starting from this value (should be used with resume)')
parser.add_argument('--max_size', type=int, help='max length of label')
parser.add_argument('--lr', type=float, help='initial learning rate')
parser.add_argument('--optimizer', type=str, help='The optimizer want to use')
parser.add_argument('--thresh', type=float, help='The threshold to replace it in the representers')
parser.add_argument('--verbose', action='store_true', help='show verbose info')
parser.add_argument('--visualize', action='store_true', help='visualize maps in tensorboard')
parser.add_argument('--force_reload', action='store_true', dest='force_reload', help='Force reload data meta')
parser.add_argument('--no-force_reload', action='store_false', dest='force_reload', help='Force reload data meta')
parser.add_argument('--validate', action='store_true', dest='validate', help='Validate during training')
parser.add_argument('--no-validate', action='store_false', dest='validate', help='Validate during training')
parser.add_argument('--print-config-only', action='store_true', help='print config without actual training')
parser.add_argument('--debug', action='store_true', dest='debug',
help='Run with debug mode, which hacks dataset num_samples to toy number')
parser.add_argument('--no-debug', action='store_false', dest='debug', help='Run without debug mode')
parser.add_argument('--benchmark', action='store_true', dest='benchmark', help='Open cudnn benchmark mode')
parser.add_argument('--no-benchmark', action='store_false', dest='benchmark', help='Turn cudnn benchmark mode off')
parser.add_argument('-d', '--distributed', action='store_true', dest='distributed', help='Use distributed training')
parser.add_argument('--local_rank', dest='local_rank', default=0, type=int, help='Use distributed training')
parser.add_argument('-gpu_num', '--num_gpus', dest='num_gpus', default=1, type=int,
help='The number of accessible gpus')
parser.set_defaults(debug=False)
parser.set_defaults(benchmark=True)
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
if args['distributed']:
torch.cuda.set_device(args['local_rank'])
torch.distributed.init_process_group(backend='nccl', init_method='env://')
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
experiment = Configurable.construct_class_from_config(experiment_args)
if not args['print_config_only']:
torch.backends.cudnn.benchmark = args['benchmark']
trainer = Trainer(experiment)
trainer.train()
def main():
parser = argparse.ArgumentParser(description='Text Recognition Training')
parser.add_argument('exp', type=str)
parser.add_argument('--resume', type=str, help='Resume from checkpoint')
parser.add_argument('--image_path', type=str, help='image path')
parser.add_argument('--result_dir', type=str, default='./demo_results/', help='path to save results')
parser.add_argument('--data', type=str,
help='The name of dataloader which will be evaluated on.')
parser.add_argument('--image_short_side', type=int, default=736,
help='The threshold to replace it in the representers')
parser.add_argument('--thresh', type=float,
help='The threshold to replace it in the representers')
parser.add_argument('--box_thresh', type=float, default=0.6,
help='The threshold to replace it in the representers')
parser.add_argument('--visualize', action='store_true',
help='visualize maps in tensorboard')
parser.add_argument('--resize', action='store_true',
help='resize')
parser.add_argument('--polygon', action='store_true',
help='output polygons if true')
parser.add_argument('--eager', '--eager_show', action='store_true', dest='eager_show',
help='Show images eagerly')
parser.add_argument('--sort_boxes', action='store_true', dest='sort_boxes',
help='Sort boxes for further works')
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
# Delete train settings, prevent of reading training dataset
experiment_args.pop('train')
experiment_args.pop('evaluation')
experiment_args.pop('validation')
experiment = Configurable.construct_class_from_config(experiment_args)
demo_handler = Demo(experiment, experiment_args, cmd=args)
if os.path.isdir(args['image_path']):
img_cnt = len(os.listdir(args['image_path']))
for idx, img in enumerate(os.listdir(args['image_path'])):
if os.path.splitext(img)[1].lower() not in ['.jpg', '.tif', '.png', '.jpeg']:
continue
t = time.time()
demo_handler.inference(os.path.join(args['image_path'], img), args['visualize'])
print("{}/{} elapsed time : {:.4f}s".format(idx + 1, img_cnt, time.time() - t))
else:
t = time.time()
demo_handler.inference(args['image_path'], args['visualize'])
print("elapsed time : {}s".format(time.time() - t))
def main():
parser = argparse.ArgumentParser(description='Convert model to ONNX')
parser.add_argument('exp', type=str)
parser.add_argument('resume', type=str, help='Resume from checkpoint')
parser.add_argument('output', type=str, help='Output ONNX path')
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
experiment = Configurable.construct_class_from_config(experiment_args)
Demo(experiment, experiment_args, cmd=args).inference()
def __init__(self, args=params):
self.RGB_MEAN = np.array([122.67891434, 116.66876762, 104.00698793])
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
self.experiment = Configurable.construct_class_from_config(
experiment_args)
self.experiment.load('evaluation', **args)
self.args = args
self.structure = self.experiment.structure
self.model_path = args['resume']
self.init_torch_tensor()
self.model = self.init_model()
self.resume(self.model, self.model_path)
self.model.eval()
def main():
parser = argparse.ArgumentParser(description='Text Recognition Training')
parser.add_argument('exp', type=str)
parser.add_argument('--batch_size',
type=int,
help='Batch size for training')
parser.add_argument('--resume', type=str, help='Resume from checkpoint')
parser.add_argument('--result_dir',
type=str,
default='./results/',
help='path to save results')
parser.add_argument('--epochs', type=int, help='Number of training epochs')
parser.add_argument(
'--start_iter',
type=int,
help=
'Begin counting iterations starting from this value (should be used with resume)'
)
parser.add_argument(
'--start_epoch',
type=int,
help=
'Begin counting epoch starting from this value (should be used with resume)'
)
parser.add_argument('--max_size', type=int, help='max length of label')
parser.add_argument(
'--data',
type=str,
help='The name of dataloader which will be evaluated on.')
parser.add_argument('--thresh',
type=float,
help='The threshold to replace it in the representers')
parser.add_argument('--box_thresh',
type=float,
default=0.6,
help='The threshold to replace it in the representers')
parser.add_argument('--verbose',
action='store_true',
help='show verbose info')
parser.add_argument('--no-verbose',
action='store_true',
help='show verbose info')
parser.add_argument('--visualize',
action='store_true',
help='visualize maps in tensorboard')
parser.add_argument('--resize', action='store_true', help='resize')
parser.add_argument('--polygon',
action='store_true',
help='output polygons if true')
parser.add_argument('--eager',
'--eager_show',
action='store_true',
dest='eager_show',
help='Show iamges eagerly')
parser.add_argument('--speed',
action='store_true',
dest='test_speed',
help='Test speed only')
parser.add_argument(
'--dest',
type=str,
help='Specify which prediction will be used for decoding.')
parser.add_argument(
'--debug',
action='store_true',
dest='debug',
help=
'Run with debug mode, which hacks dataset num_samples to toy number')
parser.add_argument('--no-debug',
action='store_false',
dest='debug',
help='Run without debug mode')
parser.add_argument('-d',
'--distributed',
action='store_true',
dest='distributed',
help='Use distributed training')
parser.add_argument('--local_rank',
dest='local_rank',
default=0,
type=int,
help='Use distributed training')
parser.add_argument('-g',
'--num_gpus',
dest='num_gpus',
default=1,
type=int,
help='The number of accessible gpus')
parser.set_defaults(debug=False, verbose=False)
args = parser.parse_args()
args = vars(args)
args = {k: v for k, v in args.items() if v is not None}
conf = Config()
experiment_args = conf.compile(conf.load(args['exp']))['Experiment']
experiment_args.update(cmd=args)
experiment = Configurable.construct_class_from_config(experiment_args)
Eval(experiment, experiment_args, cmd=args,
verbose=args['verbose']).eval(args['visualize'])
img = img[np.newaxis, :]
tensor = torch.from_numpy(img).float()
tensor = tensor.to("cuda:0")
print(tensor.shape)
batch = dict()
batch['image'] = tensor
# ===> model predict
pred = model.forward(batch, training=False)
return pred
if __name__ == "__main__":
config = "/share_sdb/clj/DB2/DB/experiments/seg_detector/ic15_resnet_vd_18.yaml"
conf = Config()
experiment_args = conf.compile(conf.load(config))['Experiment']
image_path = "/share_sdb/clj/paddle/img_10.jpg"
args = {}
args["model_path"] = ""
args["image_short_side"] = ""
args["result_dir"] = "/share_sdb/clj/DB/demo_results"
args["polygon"] = False
args["paddle_model_path"] = "/share_sdb/clj/paddle/models/ch_ppocr_server_v1.1_det_train/best_accuracy"
experiment_args.update(cmd=args)
experiment = Configurable.construct_class_from_config(experiment_args)
pred = Demo(experiment, args).inference(image_path)