def create_predictor(args, mode):
if mode == "det":
model_dir = args.det_model_dir
else:
model_dir = args.rec_model_dir
if model_dir is None:
logger.info("not find {} model file path {}".format(mode, model_dir))
sys.exit(0)
model_file_path = model_dir + "/model"
params_file_path = model_dir + "/params"
if not os.path.exists(model_file_path):
logger.info("not find model file path {}".format(model_file_path))
sys.exit(0)
if not os.path.exists(params_file_path):
logger.info("not find params file path {}".format(params_file_path))
sys.exit(0)
config = AnalysisConfig(model_file_path, params_file_path)
if args.use_gpu:
config.enable_use_gpu(args.gpu_mem, 0)
else:
config.disable_gpu()
config.set_cpu_math_library_num_threads(6)
if args.enable_mkldnn:
config.enable_mkldnn()
#config.enable_memory_optim()
config.disable_glog_info()
# use zero copy
config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
config.switch_use_feed_fetch_ops(False)
predictor = create_paddle_predictor(config)
input_names = predictor.get_input_names()
input_tensor = predictor.get_input_tensor(input_names[0])
output_names = predictor.get_output_names()
output_tensors = []
for output_name in output_names:
output_tensor = predictor.get_output_tensor(output_name)
output_tensors.append(output_tensor)
return predictor, input_tensor, output_tensors
def _set_config(self):
"""
predictor config setting
"""
cpu_config = AnalysisConfig(self.default_pretrained_model_path)
cpu_config.disable_glog_info()
cpu_config.disable_gpu()
self.cpu_predictor = create_paddle_predictor(cpu_config)
try:
_places = os.environ["CUDA_VISIBLE_DEVICES"]
int(_places[0])
use_gpu = True
except:
use_gpu = False
if use_gpu:
gpu_config = AnalysisConfig(self.default_pretrained_model_path)
gpu_config.disable_glog_info()
gpu_config.enable_use_gpu(memory_pool_init_size_mb=500, device_id=0)
self.gpu_predictor = create_paddle_predictor(gpu_config)
def __init__(self, model_path=None, mode='lac', pretraining=True, use_cuda=False):
super(LAC, self).__init__()
utils.check_cuda(use_cuda)
if model_path is None:
model_path = DEFAULT_SEG if mode == 'seg' else DEFAULT_LAC
self.model_path = model_path
# 超参数
self.args = utils.DefaultArgs(model_path)
self.args.use_cuda = use_cuda
self.args.pretraining = pretraining
config = None
if self.args.init_checkpoint and os.path.isfile(os.path.join(self.args.init_checkpoint, '__model__')):
config = AnalysisConfig(self.args.init_checkpoint)
config.disable_glog_info()
if use_cuda:
self.place = fluid.CUDAPlace(
int(os.getenv('FLAGS_selected_gpus', '0')))
config.enable_use_gpu(memory_pool_init_size_mb=500,
device_id=int(
os.getenv('FLAGS_selected_gpus', '0')),
)
else:
self.place = fluid.CPUPlace()
if config is not None:
self.predictor = create_paddle_predictor(config)
self.return_tag = mode != 'seg'
# init executor
self.exe = fluid.Executor(self.place)
self.dataset = reader.Dataset(self.args)
self.custom = None
self.batch = False
def create_predictor(args):
config = AnalysisConfig(args.model_file, args.params_file)
if args.use_gpu:
config.enable_use_gpu(args.gpu_mem, 0)
else:
config.disable_gpu()
config.disable_glog_info()
config.switch_ir_optim(args.ir_optim) # default true
if args.use_tensorrt:
config.enable_tensorrt_engine(
precision_mode=AnalysisConfig.Precision.Half
if args.use_fp16 else AnalysisConfig.Precision.Float32,
max_batch_size=args.batch_size)
config.enable_memory_optim()
# use zero copy
config.switch_use_feed_fetch_ops(False)
predictor = create_paddle_predictor(config)
return predictor
def _set_config(self, pretrained_model_path):
"""
predictor config path
"""
model_file_path = os.path.join(pretrained_model_path, 'model')
params_file_path = os.path.join(pretrained_model_path, 'params')
config = AnalysisConfig(model_file_path, params_file_path)
try:
_places = os.environ["CUDA_VISIBLE_DEVICES"]
int(_places[0])
use_gpu = True
except:
use_gpu = False
if use_gpu:
config.enable_use_gpu(8000, 0)
else:
config.disable_gpu()
if self.enable_mkldnn:
# cache 10 different shapes for mkldnn to avoid memory leak
config.set_mkldnn_cache_capacity(10)
config.enable_mkldnn()
config.disable_glog_info()
config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
config.switch_use_feed_fetch_ops(False)
predictor = create_paddle_predictor(config)
input_names = predictor.get_input_names()
input_tensor = predictor.get_input_tensor(input_names[0])
output_names = predictor.get_output_names()
output_tensors = []
for output_name in output_names:
output_tensor = predictor.get_output_tensor(output_name)
output_tensors.append(output_tensor)
return predictor, input_tensor, output_tensors
def _load_model_and_set_config(self):
'''
load model from file and set analysis config
'''
if os.path.exists(os.path.join(self.model_path, self.params_file)):
config = AnalysisConfig(
os.path.join(self.model_path, "__model__"),
os.path.join(self.model_path, self.params_file))
else:
config = AnalysisConfig(os.path.join(self.model_path))
if fluid.is_compiled_with_cuda():
config.enable_use_gpu(100, 0)
else:
config.disable_gpu()
config.switch_specify_input_names(True)
config.switch_use_feed_fetch_ops(False)
config.enable_memory_optim()
config.disable_glog_info()
config.switch_ir_optim(True)
return config
def _set_config(self):
"""
predictor config setting
"""
model_file_path = os.path.join(self.infer_model_path, 'model')
params_file_path = os.path.join(self.infer_model_path, 'params')
config = AnalysisConfig(model_file_path, params_file_path)
try:
_places = os.environ["CUDA_VISIBLE_DEVICES"]
int(_places[0])
use_gpu = True
except:
use_gpu = False
if use_gpu:
config.enable_use_gpu(8000, 0)
else:
config.disable_gpu()
config.disable_glog_info()
self.predictor = create_paddle_predictor(config)
def create_predictor_rec(args, mode):
"""
create predictor for inference
:param args: params for prediction engine
:param mode: mode
:return: predictor
"""
model_dir = args.rec_model_dir
model_file_path = model_dir + "/model"
params_file_path = model_dir + "/params"
if not os.path.exists(model_file_path):
logger.info("not find model file path {}".format(model_file_path))
sys.exit(0)
if not os.path.exists(params_file_path):
logger.info("not find params file path {}".format(params_file_path))
sys.exit(0)
config = AnalysisConfig(model_file_path, params_file_path)
config.enable_use_gpu(args.gpu_mem, 0)
# config.enable_memory_optim()
config.disable_glog_info()
config.switch_use_feed_fetch_ops(True)
predictor = create_paddle_predictor(config)
input_names = predictor.get_input_names()
for name in input_names:
input_tensor = predictor.get_input_tensor(name)
output_names = predictor.get_output_names()
output_tensors = []
for output_name in output_names:
output_tensor = predictor.get_output_tensor(output_name)
output_tensors.append(output_tensor)
return predictor, input_tensor, output_tensors
def _load_model_and_set_config(self):
'''
load model from file and set analysis config
'''
if os.path.exists(os.path.join(self.model_path, self.params_file)):
config = AnalysisConfig(
os.path.join(self.model_path, self.model_file),
os.path.join(self.model_path, self.params_file))
else:
config = AnalysisConfig(os.path.join(self.model_path))
if fluid.is_compiled_with_cuda():
config.enable_use_gpu(100, 0)
else:
config.disable_gpu()
config.switch_specify_input_names(True)
config.switch_use_feed_fetch_ops(False)
config.enable_memory_optim()
config.disable_glog_info()
# TODO: set it to True after PaddleInference fix the precision error
# in CUDA11
config.switch_ir_optim(False)
return config
def _set_config(self):
"""
predictor config setting
"""
self.model_file_path = os.path.join(self.default_pretrained_model_path, '__model__')
self.params_file_path = os.path.join(self.default_pretrained_model_path, '__params__')
cpu_config = AnalysisConfig(self.model_file_path, self.params_file_path)
cpu_config.disable_glog_info()
cpu_config.disable_gpu()
self.cpu_predictor = create_paddle_predictor(cpu_config)
try:
_places = os.environ["CUDA_VISIBLE_DEVICES"]
int(_places[0])
use_gpu = True
self.place = fluid.CUDAPlace(0)
except:
use_gpu = False
self.place = fluid.CPUPlace()
if use_gpu:
gpu_config = AnalysisConfig(self.model_file_path, self.params_file_path)
gpu_config.disable_glog_info()
gpu_config.enable_use_gpu(memory_pool_init_size_mb=1000, device_id=0)
self.gpu_predictor = create_paddle_predictor(gpu_config)
def create_predictor(args, mode):
"""
create predictor for inference
:param args: params for prediction engine
:param mode: mode
:return: predictor
"""
if mode == "det":
model_dir = args.det_model_dir
elif mode == 'cls':
model_dir = args.cls_model_dir
elif mode == 'rec':
model_dir = args.rec_model_dir
else:
raise ValueError(
"'mode' of create_predictor() can only be one of ['det', 'cls', 'rec']"
)
if model_dir is None:
logger.info("not find {} model file path {}".format(mode, model_dir))
sys.exit(0)
model_file_path = model_dir + "/model"
params_file_path = model_dir + "/params"
if not os.path.exists(model_file_path):
logger.info("not find model file path {}".format(model_file_path))
sys.exit(0)
if not os.path.exists(params_file_path):
logger.info("not find params file path {}".format(params_file_path))
sys.exit(0)
config = AnalysisConfig(model_file_path, params_file_path)
if args.use_gpu:
config.enable_use_gpu(args.gpu_mem, 0)
else:
config.disable_gpu()
config.set_cpu_math_library_num_threads(6)
if args.enable_mkldnn:
# cache 10 different shapes for mkldnn to avoid memory leak
config.set_mkldnn_cache_capacity(10)
config.enable_mkldnn()
# config.enable_memory_optim()
config.disable_glog_info()
if args.use_zero_copy_run:
config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
config.switch_use_feed_fetch_ops(False)
else:
config.switch_use_feed_fetch_ops(True)
predictor = create_paddle_predictor(config)
input_names = predictor.get_input_names()
for name in input_names:
input_tensor = predictor.get_input_tensor(name)
output_names = predictor.get_output_names()
output_tensors = []
for output_name in output_names:
output_tensor = predictor.get_output_tensor(output_name)
output_tensors.append(output_tensor)
return predictor, input_tensor, output_tensors