diff --git a/research/cv/ICNet/README.md b/research/cv/ICNet/README.md
index 00d5db4beefc67c73ba6f12cd7326c88ebf42863..971fa6c5c65e37aead2bdfa36974dffbc23d01cb 100644
--- a/research/cv/ICNet/README.md
+++ b/research/cv/ICNet/README.md
@@ -149,7 +149,6 @@ keep_checkpoint_max: 10
### Pre-training
The folder Res50V1_PRE contains the scripts for pre-training and its dataset is [image net](https://image-net.org/). More details in [GENet_Res50](https://gitee.com/mindspore/models/tree/master/research/cv/GENet_Res50)
-
- Usage:
```shell
@@ -221,18 +220,18 @@ avgtime 0.19648232793807982
## Performance
-### Training Performance
+### Distributed Training Performance
|Parameter | ICNet |
| ------------------- | --------------------------------------------------------- |
|resources | Ascend 910锛汣PU 2.60GHz, 192core锛沵emory锛�755G |
|Upload date |2021.6.1 |
|mindspore version |mindspore1.2.0 |
-|training parameter |epoch=160,batch_size=32 |
+|training parameter |epoch=160,batch_size=4 |
|optimizer |SGD optimizer锛宮omentum=0.9,weight_decay=0.0001 |
|loss function |SoftmaxCrossEntropyLoss |
-|training speed | epoch time锛�285693.557 ms per step time :42.961 ms |
-|total time |about 5 hours |
+|training speed | epoch time锛�21469.152 ms(8pcs) per step time :230.851 ms(8pcs) |
+|total time |1h1m34s(8pcs) |
|Script URL | |
|Random number seed |set_seed = 1234 |
diff --git a/research/cv/ICNet/scripts/run_distribute_train8p.sh b/research/cv/ICNet/scripts/run_distribute_train8p.sh
index f1e1e4b2a87140ddd4ae539e15e9fe10f8af1ba7..9d96fca9fb24dcbdfe4f3666780def0ded3db498 100644
--- a/research/cv/ICNet/scripts/run_distribute_train8p.sh
+++ b/research/cv/ICNet/scripts/run_distribute_train8p.sh
@@ -20,7 +20,7 @@ then
echo "Usage: bash scripts/run_distribute_train8p.sh [RANK_TABLE_FILE] [PROJECT_PATH]"
echo "Please run the script as: "
echo "bash scripts/run_distribute_train8p.sh [RANK_TABLE_FILE] [PROJECT_PATH]"
- echo "for example: bash script/run_distribute_train8p.sh /absolute/path/to/RANK_TABLE_FILE /root/ICNet/"
+ echo "for example: bash scripts/run_distribute_train8p.sh /absolute/path/to/RANK_TABLE_FILE /root/ICNet/"
echo "=============================================================================================================="
exit 1
fi
diff --git a/research/cv/ICNet/scripts/run_train1p.sh b/research/cv/ICNet/scripts/run_train1p.sh
new file mode 100644
index 0000000000000000000000000000000000000000..8fdf2b1957e8b4a8275f440937b58c803f211f83
--- /dev/null
+++ b/research/cv/ICNet/scripts/run_train1p.sh
@@ -0,0 +1,52 @@
+#!/bin/bash
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+if [ $# != 2 ]
+then
+ echo "=============================================================================================================="
+ echo "Usage: bash scripts/run_train1p.sh [PROJECT_PATH] [DEVICE_ID]"
+ echo "for example: bash scripts/run_train1p.sh /root/ICNet/ 0"
+ echo "=============================================================================================================="
+ exit 1
+fi
+
+get_real_path(){
+ if [ "${1:0:1}" == "/" ]; then
+ echo "$1"
+ else
+ echo "$(realpath -m $PWD/$1)"
+ fi
+}
+
+PATH1=$(get_real_path $1)
+
+
+if [ ! -d $PATH1 ]
+then
+ echo "error: PROJECT_PATH=$PATH1 is not a directory"
+exit 1
+fi
+
+
+rm -rf LOG
+mkdir ./LOG
+export RANK_SIZE=1
+export RANK_ID=0
+export DEVICE_ID=$2
+echo "start training for rank $i, device $DEVICE_ID"
+env > env.log
+
+python3 train.py --project_path=$PATH2 > log.txt 2>&1 &
diff --git a/research/cv/ICNet/src/models/icnet_dc.py b/research/cv/ICNet/src/models/icnet_dc.py
index e0d9441c3fea193f4bec2bed28aaeb6f21a65387..6476850fbce668371b9f88d0de8261d7deb119d5 100644
--- a/research/cv/ICNet/src/models/icnet_dc.py
+++ b/research/cv/ICNet/src/models/icnet_dc.py
@@ -28,19 +28,19 @@ context.set_context(mode=context.GRAPH_MODE, device_target='Ascend')
class ICNetdc(nn.Cell):
"""Image Cascade Network"""
- def __init__(self, nclass=19, pretrained_path="", istraining=True):
+ def __init__(self, nclass=19, pretrained_path="", istraining=True, norm_layer=nn.SyncBatchNorm):
super(ICNetdc, self).__init__()
self.conv_sub1 = nn.SequentialCell(
- _ConvBNReLU(3, 32, 3, 2),
- _ConvBNReLU(32, 32, 3, 2),
- _ConvBNReLU(32, 64, 3, 2)
+ _ConvBNReLU(3, 32, 3, 2, norm_layer=norm_layer),
+ _ConvBNReLU(32, 32, 3, 2, norm_layer=norm_layer),
+ _ConvBNReLU(32, 64, 3, 2, norm_layer=norm_layer)
)
self.istraining = istraining
self.ppm = PyramidPoolingModule()
self.backbone = SegBaseModel(root=pretrained_path)
- self.head = _ICHead(nclass)
+ self.head = _ICHead(nclass, norm_layer=norm_layer)
self.loss = ICNetLoss()
@@ -73,7 +73,6 @@ class ICNetdc(nn.Cell):
outputs = output
return outputs
-
class PyramidPoolingModule(nn.Cell):
"""PPM"""
diff --git a/research/cv/ICNet/train.py b/research/cv/ICNet/train.py
index 4fca39a9f674629b6d441682addfc437aca32ee1..d14eb3407af1b4662067eba9149266e9aa69e7f7 100644
--- a/research/cv/ICNet/train.py
+++ b/research/cv/ICNet/train.py
@@ -29,7 +29,8 @@ from mindspore.train.callback import ModelCheckpoint
from mindspore.train.callback import LossMonitor
from mindspore.train.callback import TimeMonitor
-device_id = int(os.getenv('RANK_ID'))
+rank_id = int(os.getenv('RANK_ID'))
+device_id = int(os.getenv('DEVICE_ID'))
device_num = int(os.getenv('RANK_SIZE'))
context.set_context(mode=context.GRAPH_MODE, device_target='Ascend')
@@ -52,12 +53,15 @@ def train_net():
mindrecord_dir = cfg['train']["mindrecord_dir"]
mindrecord_file = os.path.join(mindrecord_dir, prefix)
dataset = create_icnet_dataset(mindrecord_file, batch_size=cfg['train']["train_batch_size_percard"],
- device_num=device_num, rank_id=device_id)
+ device_num=device_num, rank_id=rank_id)
train_data_size = dataset.get_dataset_size()
print("data_size", train_data_size)
epoch = cfg["train"]["epochs"]
- network = ICNetdc(pretrained_path=cfg["train"]["pretrained_model_path"]) # __init__
+ if device_num > 1:
+ network = ICNetdc(pretrained_path=cfg["train"]["pretrained_model_path"]) # __init__
+ else:
+ network = ICNetdc(pretrained_path=cfg["train"]["pretrained_model_path"], norm_layer=nn.BatchNorm2d)
iters_per_epoch = train_data_size
total_train_steps = iters_per_epoch * epoch