diff --git a/research/cv/osnet/README.md b/research/cv/osnet/README.md
index f449b8b59d3e2272bacffa36a5fdf093a9742ba1..4cafb342617c77b3aee80705f88497c513a1de50 100644
--- a/research/cv/osnet/README.md
+++ b/research/cv/osnet/README.md
@@ -80,8 +80,8 @@ In this project, the file organization is recommended as below:
鈹溾攢鈹€cuhk03
鈹溾攢鈹€cuhk03_release
鈹溾攢鈹€cuhk-03.mat
- 鈹溾攢鈹€cuhk03_new_protocol_config_labeled.mat
- 鈹溾攢鈹€cuhk03_new_protocol_config_detected.mat
+ 鈹溾攢鈹€cuhk03_new_protocol_config_labeled.mat
+ 鈹溾攢鈹€cuhk03_new_protocol_config_detected.mat
鈹溾攢鈹€msmt17
鈹溾攢鈹€MSMT17_V1
鈹溾攢鈹€train
@@ -92,11 +92,13 @@ In this project, the file organization is recommended as below:
鈹溾攢鈹€list_query.txt
```
+While first using the cuhk03 dataset, Please delete the splits file generated by the default match.
+
# [Features](#contents)
# [Environment Requirements](#contents)
-- Hardware锛圓scend锛�
+- Hardware锛圓scend/GPU锛�
- Prepare hardware environment with Ascend processor.
- Framework
- [MindSpore](https://www.mindspore.cn/install/en)
@@ -115,7 +117,10 @@ In this project, the file organization is recommended as below:
鈹溾攢scripts
鈹溾攢run_train_standalone_ascend.sh # launch standalone training with ascend platform(1p)
鈹溾攢run_train_distribute_ascend.sh # launch distributed training with ascend platform(8p)
- 鈹斺攢run_eval_ascend.sh # launch evaluating with ascend platform
+ 鈹溾攢run_eval_ascend.sh # launch evaluating with ascend platform
+ 鈹溾攢run_train_standalone_gpu.sh # launch standalone training with gpu platform(1p)
+ 鈹溾攢run_train_distribute_gpu.sh # launch distributed training with gpu platform(8p)
+ 鈹斺攢run_eval_gpu.sh # launch evaluating with gpu platform
鈹溾攢src
鈹溾攢cross_entropy_loss.py # cross entropy loss
鈹溾攢dataset.py # data preprocessing
@@ -143,7 +148,7 @@ data_path:/home/osnet/datasets
```bash
# distribute training example(8p)
bash run_train_distribute_ascend.sh [RANK_TABLE_FILE] [DATASET] [PRETRAINED_CKPT_PATH](optional)
-# example: bash run_distribute_train_ascend.sh ./hccl_8p.json market1501 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
+# example: bash run_train_distribute_ascend.sh ./hccl_8p.json market1501 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
# standalone training
bash run_train_standalone_ascend.sh [DATASET] [DEVICE_ID] [PRETRAINED_CKPT_PATH](optional)
@@ -162,7 +167,7 @@ bash run_eval_ascend.sh [DATASET] [CHECKPOINT_PATH] [DEVICE_ID]
> The `PRETRAINED_CKPT_PATH` should be a checkpoint saved in the training process on ascend, it will resume from the checkpoint and continue to train.
### Launch
-- Training needs to load the parameters pre-trained on ImageNet. You can download the checkpoint file on [link](https://pan.baidu.com/s/1BLf2BwFYRXwgD44zkzExRQ), the extraction code is `1961`. After downloading, put it in the `./model_utils` folder. You can also download the `.pth` file pre-trained under Pytorch here, and convert it to `.ckpt` file through `./model_utils/pth_to_ckpt.py`.
+- Training needs to load the parameters pre-trained on ImageNet. You can download the checkpoint file on [init_osnet.ckpt](https://pan.baidu.com/s/1BLf2BwFYRXwgD44zkzExRQ), the extraction code is `1961`. After downloading, put it in the `./model_utils` folder. You can also download the [osnet_x1_0_imagenet.pth](https://drive.google.com/file/d/1LaG1EJpHrxdAxKnSCJ_i0u-nbxSAeiFY/view) file pre-trained under ImageNet via [osnet (pytorch)](https://github.com/KaiyangZhou/deep-person-reid) here, and convert it to `.ckpt` file through `./model_utils/pth_to_ckpt.py`.
- Running on local server
@@ -174,13 +179,26 @@ bash run_eval_ascend.sh [DATASET] [CHECKPOINT_PATH] [DEVICE_ID]
Ascend:
# distribute training example(8p)
bash run_train_distribute_ascend.sh [RANK_TABLE_FILE] [DATASET] [PRETRAINED_CKPT_PATH](optional)
- # example: bash run_distribute_train_ascend.sh ./hccl_8p.json market1501 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
+ # example: bash run_train_distribute_ascend.sh ./hccl_8p.json market1501 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
# standalone training
bash run_train_standalone_ascend.sh [DATASET] [DEVICE_ID] [PRETRAINED_CKPT_PATH](optional)
# example: bash run_train_standalone_ascend.sh market1501 0 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
```
+ ```bash
+ # training example
+ shell:
+ GPU:
+ # distribute training example(8p)
+ bash run_train_distribute_gpu.sh [DEVICE_NUM] [DATASET] [PRETRAINED_CKPT_PATH](optional)
+ # example: bash run_train_distribute_gpu.sh 8 market1501 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
+
+ # standalone training
+ bash run_train_standalone_gpu.sh [DATASET] [DEVICE_ID] [PRETRAINED_CKPT_PATH](optional)
+ # example: bash run_train_standalone_gpu.sh market1501 0 /home/osnet/checkpoint/market1501/osnet-240_101.ckpt
+ ```
+
- Running on ModelArts
- ModelArts (If you want to run in modelarts, please check the official documentation of [modelarts](https://support.huaweicloud.com/modelarts/), and you can start training as follows)
@@ -295,6 +313,14 @@ You can start evaluating using python or shell scripts. The usage of shell scrip
# example: bash run_eval_ascend.sh market1501 /home/osnet/scripts/output/checkpoint/market1501/osnet-240_101.ckpt 0
```
+ ```bash
+ # eval example
+ shell:
+ GPU:
+ bash run_eval_gpu.sh [DATASET] [CHECKPOINT_PATH] [DEVICE_ID]
+ # example: bash run_eval_gpu.sh market1501 /home/osnet/scripts/output/checkpoint/market1501/osnet-240_101.ckpt 0
+ ```
+
- Running on ModelArts.
```bash
@@ -381,89 +407,117 @@ Rank-20 : 97.4%
#### OSNet train on Market1501
-| Parameters | OSNet |
-| -------------------------- | ----------------------------------------------------------- |
-| Resource | Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8 |
-| uploaded Date | 18/12/2021 (month/day/year) |
-| MindSpore Version | 1.5.0 |
-| Dataset | Market1501 |
-| Training Parameters | epoch=250, batch_size = 128, lr=0.001 |
-| Optimizer | Adam |
-| Loss Function | Label Smoothing Cross Entropy Loss |
-| outputs | probability |
-| Speed | 1pc: 175.741 ms/step; 8pcs: 181.027 ms/step | | |
-| Checkpoint for Fine tuning | 29.4M (.ckpt file) |
+| Parameters | Ascend | GPU |
+| -------------------------- | ----------------------------------------------------------- | ----------------------------------------------------------- |
+| Resource | Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8 | RTX 3090; CPU 2.90GHz, 16cores; Memory 24G |
+| uploaded Date | 18/12/2021 (month/day/year) | 27/1/2022 (month/day/year) |
+| MindSpore Version | 1.5.0 | 1.5.0 |
+| Dataset | Market1501 | Market1501 |
+| Training Parameters | epoch=250, batch_size = 128, lr=0.001 | epoch=250, batch_size = 128, lr=0.001 |
+| Optimizer | Adam | Adam |
+| Loss Function | Label Smoothing Cross Entropy Loss | Label Smoothing Cross Entropy Loss |
+| outputs | probability | probability |
+| Speed | 1pc: 175.741 ms/step; 8pcs: 181.027 ms/step | 1pc: 198.320 ms/step; 8pcs: 185.140 ms/step |
+| Checkpoint for Fine tuning | 29.4M (.ckpt file) | 30.91M (.ckpt file) |
#### OSNet train on DukeMTMC-reID
-| Parameters | OSNet |
-| -------------------------- | ----------------------------------------------------------- |
-| Resource | Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8 |
-| uploaded Date | 18/12/2021 (month/day/year) |
-| MindSpore Version | 1.5.0 |
-| Dataset | DukeMTMC-reID |
-| Training Parameters | epoch=250, batch_size = 128, lr=0.001 |
-| Optimizer | Adam |
-| Loss Function | Label Smoothing Cross Entropy Loss |
-| outputs | probability |
-| Speed | 1pc: 175.904 ms/step; 8pcs: 180.340 ms/step | | |
-| Checkpoint for Fine tuning | 29.11M (.ckpt file) |
+| Parameters | Ascend | GPU |
+| -------------------------- | ----------------------------------------------------------- | ----------------------------------------------------------- |
+| Resource | Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8 | RTX 3090; CPU 2.90GHz, 16cores; Memory 24G |
+| uploaded Date | 18/12/2021 (month/day/year) | 27/1/2022 (month/day/year) |
+| MindSpore Version | 1.5.0 | 1.5.0 |
+| Dataset | DukeMTMC-reID | DukeMTMC-reID |
+| Training Parameters | epoch=250, batch_size = 128, lr=0.001 | epoch=250, batch_size = 128, lr=0.001 |
+| Optimizer | Adam | Adam |
+| Loss Function | Label Smoothing Cross Entropy Loss | Label Smoothing Cross Entropy Loss
+| outputs | probability | probability |
+| Speed | 1pc: 175.904 ms/step; 8pcs: 180.340 ms/step | 1pc: 228.166 ms/step; 8pcs: 178.790 ms/step |
+| Checkpoint for Fine tuning | 29.11M (.ckpt file) | 30.63M (.ckpt file) |
#### OSNet train on MSMT17
-| Parameters | OSNet |
+| Parameters | Ascend | GPU |
+| -------------------------- | ----------------------------------------------------------- | ----------------------------------------------------------- |
+| Resource | Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8 | RTX 3090; CPU 2.90GHz, 16cores; Memory 24G |
+| uploaded Date | 18/12/2021 (month/day/year) | 27/1/2022 (month/day/year) |
+| MindSpore Version | 1.5.0 | 1.5.0 |
+| Dataset | MSMT17 | MSMT17 |
+| Training Parameters | epoch=250, batch_size = 128, lr=0.001 | epoch=250, batch_size = 128, lr=0.001 |
+| Optimizer | Adam | Adam |
+| Loss Function | Label Smoothing Cross Entropy Loss | Label Smoothing Cross Entropy Loss |
+| outputs | probability | probability |
+| Speed | 1pc: 183.783 ms/step; 8pcs: 180.458 ms/step | 1pc: 309.691 ms/step; 8pcs: 218.579 ms/step|
+| Checkpoint for Fine tuning | 31.12M (.ckpt file) | 32.82M (.ckpt file) |
+
+#### OSNet train on CUHK03
+
+| Parameters | GPU |
| -------------------------- | ----------------------------------------------------------- |
-| Resource | Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8 |
-| uploaded Date | 18/12/2021 (month/day/year) |
+| Resource | RTX 3090; CPU 2.90GHz, 16cores; Memory 24G |
+| uploaded Date | 27/1/2022 (month/day/year) |
| MindSpore Version | 1.5.0 |
-| Dataset | MSMT17 |
+| Dataset | CUHK03 |
| Training Parameters | epoch=250, batch_size = 128, lr=0.001 |
| Optimizer | Adam |
| Loss Function | Label Smoothing Cross Entropy Loss |
| outputs | probability |
-| Speed | 1pc: 183.783 ms/step; 8pcs: 180.458 ms/step | | |
-| Checkpoint for Fine tuning | 31.12M (.ckpt file) |
+| Speed | 1pc: 144.631 ms/step; 8pcs: 237.241 ms/step |
+| Checkpoint for Fine tuning | 30.96M (.ckpt file)
### Inference Performance
#### OSNet on Market1501
-| Parameters | Ascend |
-| ------------------- | --------------------------- |
-| Resource | Ascend 910; OS Euler2.8 |
-| Uploaded Date | 18/12/2021 (month/day/year) |
-| MindSpore Version | 1.5.0 |
-| Dataset | Market1501 |
-| batch_size | 300 |
-| outputs | probability |
-| mAP | 1pc: 82.4%; 8pcs:83.7% |
-| Rank-1 | 1pc: 93.3%; 8pcs:93.9% |
+| Parameters | Ascend |GPU |
+| ------------------- | --------------------------- |--------------------------- |
+| Resource | Ascend 910; OS Euler2.8 | RTX 3090 |
+| Uploaded Date | 18/12/2021 (month/day/year) | 27/1/2022 (month/day/year) |
+| MindSpore Version | 1.5.0 | 1.5.0 |
+| Dataset | Market1501 | Market1501 |
+| batch_size | 300 | 300 |
+| outputs | probability | probability |
+| mAP | 1pc: 82.4%; 8pcs:83.7% | 1pc: 81.9%; 8pcs:83.4% |
+| Rank-1 | 1pc: 93.3%; 8pcs:93.9% | 1pc: 93.3%; 8pcs:93.8% |
#### OSNet on DukeMTMC-reID
-| Parameters | Ascend |
-| ------------------- | --------------------------- |
-| Resource | Ascend 910; OS Euler2.8 |
-| Uploaded Date | 18/12/2021 (month/day/year) |
-| MindSpore Version | 1.5.0 |
-| Dataset | DukeMTMC-reID |
-| batch_size | 300 |
-| outputs | probability |
-| mAP | 1pc: 69.8%; 8pcs:74.6% |
-| Rank-1 | 1pc: 86.2%; 8pcs:89.2% |
+| Parameters | Ascend | GPU |
+| ------------------- | --------------------------- |--------------------------- |
+| Resource | Ascend 910; OS Euler2.8 | RTX 3090 |
+| Uploaded Date | 18/12/2021 (month/day/year) | 27/1/2022 (month/day/year) |
+| MindSpore Version | 1.5.0 | 1.5.0 |
+| Dataset | DukeMTMC-reID | DukeMTMC-reID |
+| batch_size | 300 | 300 |
+| outputs | probability | probability |
+| mAP | 1pc: 69.8%; 8pcs:74.6% | 1pc: 69.8%; 8pcs:74.5% |
+| Rank-1 | 1pc: 86.2%; 8pcs:89.2% | 1pc: 86.9%; 8pcs:88.5% |
#### OSNet on MSMT17
-| Parameters | Ascend |
+| Parameters | Ascend | GPU |
+| ------------------- | --------------------------- | --------------------------- |
+| Resource | Ascend 910; OS Euler2.8 | RTX 3090 |
+| Uploaded Date | 18/12/2021 (month/day/year) | 27/1/2022 (month/day/year) |
+| MindSpore Version | 1.5.0 | 1.5.0 |
+| Dataset | MSMT17 | MSMT17 |
+| batch_size | 300 | 300 |
+| outputs | probability | probability |
+| mAP | 1pc: 43.1%; 8pcs:50.0% | 1pc: 46.5%; 8pcs:53.5% |
+| Rank-1 | 1pc: 71.5%; 8pcs:77.7% | 1pc: 71.7%; 8pcs:77.5% |
+
+#### OSNet on CUHK03
+
+| Parameters | GPU |
| ------------------- | --------------------------- |
-| Resource | Ascend 910; OS Euler2.8 |
-| Uploaded Date | 18/12/2021 (month/day/year) |
+| Resource | RTX 3090 |
+| Uploaded Date | 27/1/2022 (month/day/year) |
| MindSpore Version | 1.5.0 |
-| Dataset | MSMT17 |
+| Dataset | CUHK03 |
| batch_size | 300 |
| outputs | probability |
-| mAP | 1pc: 43.1%; 8pcs:50.0% |
-| Rank-1 | 1pc: 71.5%; 8pcs:77.7% |
+| mAP | 1pc: 60.8%; 8pcs:54.9% |
+| Rank-1 | 1pc: 65.4%; 8pcs:60.1% |
# [Description of Random Situation](#contents)
diff --git a/research/cv/osnet/eval.py b/research/cv/osnet/eval.py
index 7336971b797327278f53105c2e66ef60bf255b9c..69c0bd9e9ef3e3c2f842cf72f2be7bb4c4737ace 100644
--- a/research/cv/osnet/eval.py
+++ b/research/cv/osnet/eval.py
@@ -179,4 +179,9 @@ def eval_net(net=None):
i += 1
if __name__ == '__main__':
+ if config.target == 'msmt17' or config.target == 'cuhk03 ':
+ config.dist_metric = 'cosine'
+ else:
+ config.dist_metric = 'euclidean'
+ print(config.dist_metric)
eval_net()
diff --git a/research/cv/osnet/export.py b/research/cv/osnet/export.py
index a69caf0ec71ab9a4e64eafda4daaa04187d0140c..49b46560c5f1a34add718e90b6d3ef547261de10 100644
--- a/research/cv/osnet/export.py
+++ b/research/cv/osnet/export.py
@@ -24,7 +24,7 @@ from mindspore import Tensor, load_checkpoint, load_param_into_net, export, cont
from src.osnet import create_osnet
-from src.datasets_define import (Market1501, DukeMTMCreID, MSMT17, CUHK03)
+from src.datasets_define import Market1501, DukeMTMCreID, MSMT17, CUHK03
from model_utils.config import config
from model_utils.moxing_adapter import moxing_wrapper
from model_utils.device_adapter import get_device_id
diff --git a/research/cv/osnet/model_utils/transforms.py b/research/cv/osnet/model_utils/transforms.py
index 2f6a849bec59a5f612060c936028f6c5c275eaaf..779d5e89f21ff81f237d3a2a9f9c6d3c872a5271 100644
--- a/research/cv/osnet/model_utils/transforms.py
+++ b/research/cv/osnet/model_utils/transforms.py
@@ -17,7 +17,7 @@
import math
import random
-from mindspore.dataset.vision.c_transforms import(Resize, Rescale, Normalize, HWC2CHW, RandomHorizontalFlip)
+from mindspore.dataset.vision.c_transforms import Resize, Rescale, Normalize, HWC2CHW, RandomHorizontalFlip
from mindspore.dataset.transforms.c_transforms import Compose
diff --git a/research/cv/osnet/osnet_config.yaml b/research/cv/osnet/osnet_config.yaml
index 5fca76f9d0d45c8e484f1cc450e683191d352564..5479085e88a42750b5692d01bffad687eecd695e 100644
--- a/research/cv/osnet/osnet_config.yaml
+++ b/research/cv/osnet/osnet_config.yaml
@@ -35,6 +35,7 @@ save_checkpoint: True
save_checkpoint_epochs: 10
pretrained_dir: 'model_utils'
run_distribute: False
+LogSummary: False
# optimizer
adam_beta1: 0.9
diff --git a/research/cv/osnet/scripts/run_eval_gpu.sh b/research/cv/osnet/scripts/run_eval_gpu.sh
new file mode 100644
index 0000000000000000000000000000000000000000..03fd17828001f1575de72dfe0e8cd9701609afb1
--- /dev/null
+++ b/research/cv/osnet/scripts/run_eval_gpu.sh
@@ -0,0 +1,58 @@
+#!/bin/bash
+# Copyright 2022 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 WARRANT IES OR CONITTONS OF ANY KIND锛� either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ====================================================================================
+
+if [ $# != 3 ]
+then
+ echo "Usage: bash run_eval_ascend.sh [market1501|dukemtmcreid|cuhk03|msmt17] [CHECKPOINT_PATH] [DEVICE_ID]"
+exit 1
+fi
+
+if [ $1 != "market1501" ] && [ $1 != "dukemtmcreid" ] && [ $1 != "cuhk03" ] && [ $1 != "msmt17" ]
+then
+ echo "error: the selected dataset is not market1501, dukemtmcreid, cuhk03 or msmt17"
+exit 1
+fi
+dataset_name=$1
+
+
+if [ ! -f $2 ]
+then
+ echo "error: CHECKPOINT_PATH=$2 is not a file"
+exit 1
+fi
+PATH1=$(realpath $2)
+
+BASEPATH=$(cd "`dirname $0`" || exit; pwd)
+export PYTHONPATH=${BASEPATH}:$PYTHONPATH
+
+ulimit -u unlimited
+export DEVICE_NUM=1
+export DEVICE_ID=$3
+
+export RANK_SIZE=1
+export RANK_ID=$3
+
+config_path="${BASEPATH}/../osnet_config.yaml"
+echo "config path is : ${config_path}"
+
+if [ -d "./eval" ];
+then
+ rm -rf ./eval
+fi
+mkdir ./eval
+cd ./eval || exit
+echo "start evaluating for device $DEVICE_ID"
+python ${BASEPATH}/../eval.py --config_path=$config_path --checkpoint_file_path=$PATH1 --device_target="GPU" --target=$dataset_name> ./eval.log 2>&1 &
diff --git a/research/cv/osnet/scripts/run_train_distribute_gpu.sh b/research/cv/osnet/scripts/run_train_distribute_gpu.sh
new file mode 100644
index 0000000000000000000000000000000000000000..7bc0d56fa92438515897b57831ae6b341feeec36
--- /dev/null
+++ b/research/cv/osnet/scripts/run_train_distribute_gpu.sh
@@ -0,0 +1,74 @@
+#!/bin/bash
+# Copyright 2022 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 WARRANT IES OR CONITTONS OF ANY KIND锛� either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ====================================================================================
+
+if [ $# != 2 ] && [ $# != 3 ]
+then
+ echo "Usage: sh run_train_distribute_ascend.sh [DEVICE_NUM] [market1501|dukemtmcreid|cuhk03|msmt17] [PRETRAINED_CKPT_PATH](optional)"
+exit 1
+fi
+
+if [ $2 != "market1501" ] && [ $2 != "dukemtmcreid" ] && [ $2 != "cuhk03" ] && [ $2 != "msmt17" ]
+then
+ echo "error: the selected dataset is not market1501, dukemtmcreid, cuhk03 or msmt17"
+exit 1
+fi
+dataset_name=$2
+
+if [ $# == 3 ]
+then
+ if [ ! -f $3 ]
+ then
+ echo "error: PRETRAINED_CKPT_PATH=$3 is not a file"
+ exit 1
+ fi
+ PATH1=$(realpath $3)
+fi
+
+ulimit -u unlimited
+export RANK_SIZE=$1
+
+
+BASEPATH=$(cd "`dirname $0`" || exit; pwd)
+config_path="${BASEPATH}/../osnet_config.yaml"
+echo "config path is : ${config_path}"
+if [ -d "./train" ];
+then
+ rm -rf ./train
+fi
+mkdir ./train
+cd ./train || exit
+echo "start training on $RANK_SIZE device"
+
+if [ $# == 2 ];
+then
+ mpirun -n $RANK_SIZE --output-filename log_output --merge-stderr-to-stdout --allow-run-as-root\
+ python ${BASEPATH}/../train.py \
+ --config_path=$config_path \
+ --device_target="GPU" \
+ --run_distribute=True \
+ --source=$dataset_name \
+ --output_path='./osnet_ckpt_output' > train.log 2>&1 &
+else
+ mpirun -n $RANK_SIZE --output-filename log_output --merge-stderr-to-stdout --allow-run-as-root\
+ python ${BASEPATH}/../train.py \
+ --config_path=$config_path \
+ --device_target="GPU" \
+ --run_distribute=True \
+ --source=$dataset_name \
+ --output_path='./osnet_ckpt_output' \
+ --checkpoint_file_path=$PATH1> train.log 2>&1 &
+fi
+
diff --git a/research/cv/osnet/scripts/run_train_standalone_gpu.sh b/research/cv/osnet/scripts/run_train_standalone_gpu.sh
new file mode 100644
index 0000000000000000000000000000000000000000..5838576f58003b6dfc281c07469fe9ba69f88a76
--- /dev/null
+++ b/research/cv/osnet/scripts/run_train_standalone_gpu.sh
@@ -0,0 +1,65 @@
+#!/bin/bash
+# Copyright 2022 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 WARRANT IES OR CONITTONS OF ANY KIND锛� either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ====================================================================================
+
+if [ $# != 2 ] && [ $# != 3 ]
+then
+ echo "Usage: sh run_train_standalone_ascend.sh [market1501|dukemtmcreid|cuhk03|msmt17] [DEVICE_ID] [PRETRAINED_CKPT_PATH](optional)"
+exit 1
+fi
+
+if [ $1 != "market1501" ] && [ $1 != "dukemtmcreid" ] && [ $1 != "cuhk03" ] && [ $1 != "msmt17" ]
+then
+ echo "error: the selected dataset is not market1501, dukemtmcreid, cuhk03 or msmt17"
+exit 1
+fi
+dataset_name=$1
+
+if [ $# == 3 ]
+then
+ if [ ! -f $3 ]
+ then
+ echo "error: PRETRAINED_CKPT_PATH=$3 is not a file"
+ exit 1
+ fi
+ PATH1=$(get_real_path $3)
+fi
+
+ulimit -u unlimited
+export DEVICE_NUM=1
+export DEVICE_ID=$2
+
+export RANK_SIZE=1
+export RANK_ID=$2
+
+BASEPATH=$(cd "`dirname $0`" || exit; pwd)
+config_path="${BASEPATH}/../osnet_config.yaml"
+echo "config path is : ${config_path}"
+
+if [ -d "./train" ];
+then
+ rm -rf ./train
+fi
+mkdir ./train
+cd ./train || exit
+echo "start training for device $DEVICE_ID"
+if [ $# == 2 ];
+then
+ python ${BASEPATH}/../train.py --config_path=$config_path --source=$dataset_name \
+ --device_target="GPU" --output_path='./output'> train.log 2>&1 &
+else
+ python ${BASEPATH}/../train.py --config_path=$config_path --source=$dataset_name \
+ --device_target="GPU" --output_path='./output' --checkpoint_file_path=$PATH1> train.log 2>&1 &
+fi
diff --git a/research/cv/osnet/src/dataset.py b/research/cv/osnet/src/dataset.py
index 3dabf1eaf5bb5a64f20ada7cc878a8062830341a..4b121487421b6efbcd743aff4d6de5ab9bb71297 100644
--- a/research/cv/osnet/src/dataset.py
+++ b/research/cv/osnet/src/dataset.py
@@ -18,7 +18,7 @@
import os
import mindspore.dataset as ds
from model_utils.transforms import build_train_transforms, build_test_transforms
-from .datasets_define import (Market1501, DukeMTMCreID, MSMT17, CUHK03)
+from .datasets_define import Market1501, DukeMTMCreID, MSMT17, CUHK03
def init_dataset(name, **kwargs):
diff --git a/research/cv/osnet/train.py b/research/cv/osnet/train.py
index 3312aed65e2af36a420be73cb7d8fdcd0bb959f6..8980226358f6a2b67b8c4704babdfb9932d8c4d0 100644
--- a/research/cv/osnet/train.py
+++ b/research/cv/osnet/train.py
@@ -1,4 +1,4 @@
-# Copyright 2021 Huawei Technologies Co., Ltd
+# Copyright 2022 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.
@@ -27,8 +27,8 @@ import mindspore.ops as ops
from mindspore.common import set_seed
from mindspore import Tensor, Model, context
from mindspore import load_checkpoint, load_param_into_net
-from mindspore.train.callback import (ModelCheckpoint, CheckpointConfig, LossMonitor,
- TimeMonitor, SummaryCollector)
+from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor,\
+ TimeMonitor, SummaryCollector
from mindspore.train.loss_scale_manager import FixedLossScaleManager
from mindspore.communication.management import init, get_rank
@@ -39,7 +39,7 @@ from src.lr_generator import step_lr, multi_step_lr
from src.cross_entropy_loss import CrossEntropyLoss
from model_utils.config import config
from model_utils.moxing_adapter import moxing_wrapper
-from model_utils.device_adapter import get_device_id, get_device_num, get_rank_id
+from model_utils.device_adapter import get_device_id, get_device_num
set_seed(1)
@@ -53,6 +53,10 @@ class LossCallBack(LossMonitor):
super(LossCallBack, self).__init__()
self.has_trained_epoch = has_trained_epoch
+ def begin(self, run_context):
+ cb_params = run_context.original_args()
+ cb_params.init_time = time.time()
+
def step_end(self, run_context):
'''check loss at the end of each step.'''
cb_params = run_context.original_args()
@@ -74,6 +78,11 @@ class LossCallBack(LossMonitor):
print("epoch: %s step: %s, loss is %s" % (cb_params.cur_epoch_num + int(self.has_trained_epoch),
cur_step_in_epoch, loss), flush=True)
+ def end(self, run_context):
+ cb_params = run_context.original_args()
+ end_time = time.time()
+ print("total_time:", (end_time-cb_params.init_time)*1000, "ms")
+
def init_lr(num_batches):
'''initialize learning rate.'''
@@ -96,7 +105,7 @@ def check_isfile(fpath):
def load_from_checkpoint(net):
- '''load parameters when resuming from a checkpoint for training.'''
+ '''load parameters when resuming from a checkpoints for training.'''
param_dict = load_checkpoint(config.checkpoint_file_path)
if param_dict:
if param_dict.get("epoch_num") and param_dict.get("step_num"):
@@ -114,7 +123,7 @@ def set_save_ckpt_dir():
"""set save ckpt dir"""
ckpt_save_dir = os.path.join(config.output_path, config.checkpoint_path, config.source)
if config.enable_modelarts and config.run_distribute:
- ckpt_save_dir = ckpt_save_dir + "ckpt_" + str(get_rank_id()) + "/"
+ ckpt_save_dir = ckpt_save_dir + "ckpt_" + str(get_rank()) + "/"
else:
if config.run_distribute:
ckpt_save_dir = ckpt_save_dir + "ckpt_" + str(get_rank()) + "/"
@@ -155,7 +164,7 @@ def modelarts_pre_process():
sync_lock = "/tmp/unzip_sync.lock"
# Each server contains 8 devices as most.
- if get_device_id() % min(get_device_num(), 8) == 0 and not os.path.exists(sync_lock):
+ if get_rank() % min(get_device_num(), 8) == 0 and not os.path.exists(sync_lock):
print("Zip file path: ", zip_file_1)
print("Unzip file save dir: ", save_dir_1)
unzip(zip_file_1, save_dir_1)
@@ -170,7 +179,7 @@ def modelarts_pre_process():
break
time.sleep(1)
- print("Device: {}, Finish sync unzip data from {} to {}.".format(get_device_id(), zip_file_1, save_dir_1))
+ print("Device: {}, Finish sync unzip data from {} to {}.".format(get_rank(), zip_file_1, save_dir_1))
@moxing_wrapper(pre_process=modelarts_pre_process)
@@ -178,6 +187,16 @@ def train_net():
"""train net"""
context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target)
device_num = get_device_num()
+ if config.device_target == "GPU":
+ context.set_context(enable_graph_kernel=True)
+ if device_num > 1:
+ context.reset_auto_parallel_context()
+ context.set_auto_parallel_context(device_num=device_num, parallel_mode='data_parallel',
+ gradients_mean=True)
+ init()
+ rank_id = get_rank()
+ else:
+ rank_id = 0
if config.device_target == "Ascend":
device_id = get_device_id()
context.set_context(device_id=device_id)
@@ -186,6 +205,9 @@ def train_net():
context.set_auto_parallel_context(device_num=device_num, parallel_mode='data_parallel',
gradients_mean=True)
init()
+ rank_id = get_rank()
+ else:
+ rank_id = 0
num_classes, dataset1 = dataset_creator(root=config.data_path, height=config.height, width=config.width,
@@ -202,6 +224,7 @@ def train_net():
cuhk03_classic_split=config.cuhk03_classic_split, mode='train')
num_batches = dataset1.get_dataset_size()
+
if config.checkpoint_file_path and check_isfile(config.checkpoint_file_path):
fpath = osp.abspath(osp.expanduser(config.checkpoint_file_path))
if not osp.exists(fpath):
@@ -216,7 +239,8 @@ def train_net():
crit = CrossEntropyLoss(num_classes=num_classes, label_smooth=config.label_smooth)
lr = init_lr(num_batches=num_batches)
loss_scale = FixedLossScaleManager(config.loss_scale, drop_overflow_update=False)
- summary_collector = SummaryCollector(summary_dir='./summary_dir', collect_freq=1)
+ if config.LogSummary:
+ summary_collector = SummaryCollector(summary_dir='./summary_dir', collect_freq=1)
time_cb = TimeMonitor(data_size=num_batches)
@@ -228,7 +252,10 @@ def train_net():
model1 = Model(network=net, optimizer=opt1, loss_fn=crit, amp_level="O3",
loss_scale_manager=loss_scale)
loss_cb1 = LossCallBack(has_trained_epoch=0)
- cb1 = [time_cb, loss_cb1, summary_collector]
+ if config.LogSummary:
+ cb1 = [time_cb, loss_cb1, summary_collector]
+ else:
+ cb1 = [time_cb, loss_cb1]
model1.train(config.fixbase_epoch, dataset1, cb1, dataset_sink_mode=True)
loss_cb2 = LossCallBack(config.start_epoch)
@@ -238,9 +265,13 @@ def train_net():
weight_decay=config.weight_decay, loss_scale=config.loss_scale)
model2 = Model(network=net, optimizer=opt2, loss_fn=crit, amp_level="O3",
loss_scale_manager=loss_scale)
- cb2 = [time_cb, loss_cb2, summary_collector]
+ if config.LogSummary:
+ cb2 = [time_cb, loss_cb2, summary_collector]
+ else:
+ cb2 = [time_cb, loss_cb2]
+
if config.save_checkpoint:
- if not config.run_distribute or (config.run_distribute and get_device_id() % 8 == 0):
+ if not config.run_distribute or (config.run_distribute and rank_id % 8 == 0):
ckpt_append_info = [{"epoch_num": config.start_epoch, "step_num": config.start_epoch}]
config_ck = CheckpointConfig(save_checkpoint_steps=config.save_checkpoint_epochs * num_batches,
keep_checkpoint_max=10, append_info=ckpt_append_info)