Operating Environment Requirements

Before the installation, ensure that the Linux OS version and architecture meet the requirements listed in the following table and the system memory capacity is 32 GB or above.

**Table 1** Hardware architecture and OS version
Hardware Form	OS Version
Atlas 800 training server (model 9000)	Ubuntu 18.04.1
Atlas 800 training server (model 9010)	Ubuntu 18.04.1
Atlas 800 inference server (model 3000) + Atlas 300T training card (model 9000)	Ubuntu 18.04.1
Atlas 800 inference server (model 3010) + Atlas 300T training card (model 9000)	Ubuntu 18.04.1

Query the version and architecture of the running OS.
```
uname -m && cat /etc/*release
```
Check whether the Python environment is installed.
```
python3.9 --version
```
Ensure that the CANN development kit (6.0.1) has been successfully installed in the environment. For details about the installation process and how to obtain the development kit, see the CANN Software Installation Guide.

Run the pip3 list command to check the software packages in the environment. For details, see Table 2.

**Table 2** Software packages and versions
Component	Version	Component	Version
NumPy	≥ 1.17.0, ≤ 1.23.5	EasyDict	≥ 1.9
Protobuf	≥ 3.13.0, ≤ 3.20.1	psutil	≥ 5.7.0
asttokens	≥ 1.1.13	SymPy	≥ 1.4
Pillow	== 9.2.0	PyYAML	≥ 6
SciPy	≥ 1.5.2	pathlib2	≥ 2.3.6
CFFI	≥ 1.12.3	grpcio	≥ 1.43.0
wheel	≥ 0.32.0	grpcio-tools	≥ 1.43.0
Decorator	≥ 4.4.0	Requests	≥ 2.26.0
setuptools	≥ 40.8.0	xml-python	≥ 0.4.3
Matplotlib	≥ 3.1.3	python-Levenshtein	≥ 0.12.2
OpenCV-Python	≥ 4.1.2.26	seaborn	≥ 0.11.1
scikit-learn	≥ 1.0.1	tqdm	≥ 4.62.3
pandas	≥ 1.3.5	Cython	≥ 0.29.26
astunparse	≥ 1.6.3	defusedxml	≥ 0.7.1
packaging	≥ 20.0	YAPF	≥ 0.32.0
pycocotools	≥ 2.0.3	mindspore-ascend	≥ 1.7.0
pydantic	≥ 1.8.2	-	-

Obtain the source code of dependencies.

**Table 3** Addresses for the source code and training code dependency repositories
Open Source Address	Training Code Dependency Repository
https://gitee.com/mindspore/models/tree/r1.2/official/cv/crnn/src	models-r1.2/official/cv/crnn/src
https://gitee.com/mindspore/models/tree/r1.2/official/cv/ctpn/src	models-r1.2/official/cv/ctpn/src
https://gitee.com/mindspore/models/tree/r1.2/official/cv/resnet/src	models-r1.2/official/cv/resnet/src
https://gitee.com/mindspore/models/tree/r1.2/official/cv/ssd/src	models-r1.2/official/cv/ssd/src
https://gitee.com/mindspore/models/tree/r1.2/official/cv/yolov4/src	models-r1.2/official/cv/yolov4/src
https://gitee.com/mindspore/models/tree/r1.2/official/cv/unet/src	models-r1.2/official/cv/unet/src
https://gitee.com/mindspore/models/blob/r1.5/official/cv/resnet/src/metric.py	models-r1.2/official/cv/resnet/src/metric.py
https://github.com/Sanster/tf_ctpn/blob/master/tools/ICDAR13_Det/script.py	models-r1.2/official/cv/ctpn/src/script.py
https://github.com/SakuraRiven/EAST/blob/master/evaluate/rrc_evaluation_funcs.py	models-r1.2/official/cv/ctpn/src/rrc_evaluation_funcs.py

After you obtain the source code, set the environment variable MINDSPORE_CODE_PATH to models-r1.2/official/cv.

Ensure that the permissions on all files in the models-r1.2/official/cv directory are not higher than 640.

Decompress the model package and add the suffix --no-same-owner to ensure that the permission does not change after the decompression.
```
tar -zxvf Ascend-mindxsdk-mxtraining_{version}_linux.tar.gz --no-same-owner
```

Obtain a pre-trained model. For details about how to obtain the model and the storage path of each model, see Table 4.

**Table 4** Pre-trained models and storage paths
Model	Download Link	Storage Path
SSD	Link	ssd_mobilenet_fpn_mindspore\pre_trained_ckpt ssd_tiled_dataset_mindspore\pre_trained_ckpt
CRNN	Link	crnn_mindspore\pre_trained_ckpt
ResNet50	Link	resnet50_mindspore\pre_trained_ckpt
CTPN	Link	ctpn_mindspore\pre_trained_ckpt
Yolov4	Link	yolov4_mindspore\pre_trained_ckpt

Parent topic: Training Component