This commit is contained in:
huchunmei 2021-06-16 18:39:32 +08:00
parent 1e9786e7a8
commit fa6e4dd4f9
11 changed files with 23 additions and 31 deletions

View File

@ -162,16 +162,16 @@ def modelarts_process():
print("#" * 200, os.listdir(save_dir_1))
print("#" * 200, os.listdir(os.path.join(config.data_path, config.modelarts_dataset_unzip_name)))
config.dataset_path = os.path.join(config.data_path, config.modelarts_dataset_unzip_name)
config.coco_root = config.dataset_path
config.checkpoint_path = os.path.join(config.dataset_path, config.ckpt_path)
config.ann_file = os.path.join(config.dataset_path, config.ann_file)
config.mindrecord_dir = os.path.join(config.dataset_path, config.mindrecord_dir)
config.coco_root = os.path.join(config.data_path, config.modelarts_dataset_unzip_name)
config.checkpoint_path = os.path.join(config.output_path, config.ckpt_path)
config.ann_file = os.path.join(config.coco_root, config.ann_file)
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=get_device_id())
@moxing_wrapper(pre_process=modelarts_process)
def eval_():
config.mindrecord_dir = os.path.join(config.coco_root, config.mindrecord_dir)
print('\neval.py config:\n', config)
prefix = "MaskRcnn_eval.mindrecord"
mindrecord_dir = config.mindrecord_dir
mindrecord_file = os.path.join(mindrecord_dir, prefix)

View File

@ -77,6 +77,6 @@ do
echo "start training for rank $RANK_ID, device $DEVICE_ID"
env > env.log
taskset -c $cmdopt python train.py --do_train=True --device_id=$i --rank_id=$i --run_distribute=True --device_num=$DEVICE_NUM \
--pre_trained=$PATH2 --data_path=$PATH3 &> log &
--pre_trained=$PATH2 --coco_root=$PATH3 &> log &
cd ..
done

View File

@ -65,5 +65,5 @@ cd ./eval || exit
env > env.log_eval
echo "start eval for device $DEVICE_ID"
python ./eval.py --device_id=$DEVICE_ID --ann_file=$PATH1 --checkpoint_path=$PATH2 \
--data_path=$PATH3 &> log_eval.txt &
--coco_root=$PATH3 &> log_eval.txt &
cd ..

View File

@ -56,5 +56,5 @@ cp -r ../src ./train
cd ./train || exit
echo "start training for device $DEVICE_ID"
env > env.log
python ./train.py --do_train=True --device_id=$DEVICE_ID --pre_trained=$PATH1 --data_path=$PATH2 &> log.txt &
python ./train.py --do_train=True --device_id=$DEVICE_ID --pre_trained=$PATH1 --coco_root=$PATH2 &> log.txt &
cd ..

View File

@ -31,12 +31,6 @@ from .model_utils.config import config
config.mask_shape = (28, 28)
if config.enable_modelarts and config.need_modelarts_dataset_unzip:
config.coco_root = os.path.join(config.data_path, config.modelarts_dataset_unzip_name)
else:
config.coco_root = config.data_path
config.mindrecord_dir = os.path.join(config.coco_root, config.mindrecord_dir)
def bbox_overlaps(bboxes1, bboxes2, mode='iou'):
"""Calculate the ious between each bbox of bboxes1 and bboxes2.

View File

@ -98,16 +98,17 @@ def modelarts_pre_process():
print("#" * 200, os.listdir(save_dir_1))
print("#" * 200, os.listdir(os.path.join(config.data_path, config.modelarts_dataset_unzip_name)))
config.dataset_path = os.path.join(config.data_path, config.modelarts_dataset_unzip_name)
config.pre_trained = os.path.join(config.dataset_path, config.ckpt_path)
config.coco_root = os.path.join(config.data_path, config.modelarts_dataset_unzip_name)
config.pre_trained = os.path.join(config.coco_root, config.pre_trained)
config.save_checkpoint_path = config.output_path
config.mindrecord_dir = os.path.join(config.dataset_path, config.mindrecord_dir)
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", device_id=get_device_id())
@moxing_wrapper(pre_process=modelarts_pre_process)
def train_maskrcnn():
config.mindrecord_dir = os.path.join(config.coco_root, config.mindrecord_dir)
print('\ntrain.py config:\n', config)
print("Start train for maskrcnn!")
if not config.do_eval and config.run_distribute:
init()
@ -130,12 +131,12 @@ def train_maskrcnn():
if not os.path.isdir(mindrecord_dir):
os.makedirs(mindrecord_dir)
if config.dataset == "coco":
if os.path.isdir(config.data_path):
if os.path.isdir(config.coco_root):
print("Create Mindrecord.")
data_to_mindrecord_byte_image("coco", True, prefix)
print("Create Mindrecord Done, at {}".format(mindrecord_dir))
else:
raise Exception("data_path not exits.")
raise Exception("coco_root not exits.")
else:
if os.path.isdir(config.IMAGE_DIR) and os.path.exists(config.ANNO_PATH):
print("Create Mindrecord.")

View File

@ -11,7 +11,7 @@ device_target: Ascend
enable_profiling: False
# ==============================================================================
modelarts_dataset_unzip_name: 'ImageNet_Original'
modelarts_dataset_unzip_name: 'cifar10'
need_modelarts_dataset_unzip: True
# config for mobilenet, cifar10
@ -34,11 +34,10 @@ lr_max: 0.1
# Image classification - train
dataset: 'cifar10'
run_distribute: True
run_distribute: False
device_num: 1
dataset_path: "/cache/data"
device_target: 'Ascend'
pre_trained: ""
pre_trained: ''
parameter_server: False
# Image classification - eval

View File

@ -36,11 +36,10 @@ lr_end: 0.0
# Image classification - train
dataset: 'imagenet2012'
run_distribute: True
run_distribute: False
device_num: 1
dataset_path: "/cache/data"
device_target: 'Ascend'
pre_trained: ""
pre_trained: ''
parameter_server: False
# Image classification - eval

View File

@ -97,8 +97,8 @@ def modelarts_pre_process():
print("#" * 200, os.listdir(os.path.join(config.data_path, config.modelarts_dataset_unzip_name)))
config.dataset_path = os.path.join(config.data_path, config.modelarts_dataset_unzip_name)
config.ckpt_path = config.output_path
config.pre_trained = os.path.join(config.dataset_path, config.pre_trained)
config.save_checkpoint_path = config.output_path
# config.pre_trained = os.path.join(config.dataset_path, config.pre_trained)
@moxing_wrapper(pre_process=modelarts_pre_process)

View File

@ -19,7 +19,6 @@ audio_path: "/cache/data"
npy_path: "/cache/data"
info_path: "/cache/data"
info_name: 'annotations_final.csv'
device_target: 'Ascend'
device_id: 0
mr_path: "/cache/data"
mr_name: ['train', 'val']
@ -41,7 +40,7 @@ prefix: 'MusicTagger'
model_name: 'MusicTagger-10_543.ckpt'
# export
file_name: "/cache/data/musicTagger/fcn-4.air"
file_name: "fcn-4"
file_format: "MINDIR"
# 310 infer

View File

@ -92,5 +92,5 @@ if __name__ == "__main__":
save_checkpoint_steps=config.save_step,
keep_checkpoint_max=config.keep_checkpoint_max,
prefix=config.prefix,
directory=config.checkpoint_path + "_{}".format(get_device_id()))
directory=config.checkpoint_path) # + "_{}".format(get_device_id())
print("train success")