modify export for centerface, fix yolov4 export bug

This commit is contained in:
yuzhenhua 2020-11-20 11:42:13 +08:00
parent 7689062c7d
commit 6336825d7a
2 changed files with 30 additions and 38 deletions

View File

@ -12,51 +12,43 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""Convert ckpt to air."""
import os
import argparse
import numpy as np
from mindspore import context
from mindspore import Tensor
from mindspore.train.serialization import export, load_checkpoint, load_param_into_net
import mindspore
from mindspore import context, Tensor
from mindspore.train.serialization import load_checkpoint, load_param_into_net, export
from src.centerface import CenterfaceMobilev2
from src.config import ConfigCenterface
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", save_graphs=False)
parser = argparse.ArgumentParser(description='centerface export')
parser.add_argument("--device_id", type=int, default=0, help="Device id")
parser.add_argument("--batch_size", type=int, default=1, help="batch size")
parser.add_argument("--ckpt_file", type=str, required=True, help="Checkpoint file path.")
parser.add_argument("--file_name", type=str, default="centerface.air", help="output file name.")
parser.add_argument('--file_format', type=str, choices=["AIR", "ONNX", "MINDIR"], default='AIR', help='file format')
args = parser.parse_args()
def save_air():
"""Save air file"""
print('============= centerface start save air ==================')
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=args.device_id)
parser = argparse.ArgumentParser(description='Convert ckpt to air')
parser.add_argument('--pretrained', type=str, default='', help='pretrained model to load')
parser.add_argument('--batch_size', type=int, default=8, help='batch size')
if __name__ == '__main__':
config = ConfigCenterface()
net = CenterfaceMobilev2()
args = parser.parse_args()
network = CenterfaceMobilev2()
param_dict = load_checkpoint(args.ckpt_file)
param_dict_new = {}
for key, values in param_dict.items():
if key.startswith('moments.') or key.startswith('moment1.') or key.startswith('moment2.'):
continue
elif key.startswith('centerface_network.'):
param_dict_new[key[19:]] = values
else:
param_dict_new[key] = values
if os.path.isfile(args.pretrained):
param_dict = load_checkpoint(args.pretrained)
param_dict_new = {}
for key, values in param_dict.items():
if key.startswith('moments.') or key.startswith('moment1.') or key.startswith('moment2.'):
continue
elif key.startswith('centerface_network.'):
param_dict_new[key[19:]] = values
else:
param_dict_new[key] = values
load_param_into_net(network, param_dict_new)
print('load model {} success'.format(args.pretrained))
load_param_into_net(net, param_dict_new)
net.set_train(False)
input_data = np.random.uniform(low=0, high=1.0, size=(args.batch_size, 3, 832, 832)).astype(np.float32)
tensor_input_data = Tensor(input_data)
export(network, tensor_input_data,
file_name=args.pretrained.replace('.ckpt', '_' + str(args.batch_size) + 'b.air'), file_format='AIR')
print("export model success.")
if __name__ == "__main__":
save_air()
input_data = Tensor(np.zeros([args.batch_size, 3, config.input_h, config.input_w]), mindspore.float32)
export(net, input_data, file_name=args.file_name, file_format=args.file_format)

View File

@ -26,7 +26,7 @@ parser.add_argument("--device_id", type=int, default=0, help="Device id")
parser.add_argument("--batch_size", type=int, default=1, help="batch size")
parser.add_argument("--testing_shape", type=int, default=608, help="test shape")
parser.add_argument("--ckpt_file", type=str, required=True, help="Checkpoint file path.")
parser.add_argument("--file_name", type=str, default="ssd.air", help="output file name.")
parser.add_argument("--file_name", type=str, default="yolov4.air", help="output file name.")
parser.add_argument('--file_format', type=str, choices=["AIR", "ONNX", "MINDIR"], default='AIR', help='file format')
args = parser.parse_args()