portrait_picture

中文|English

This sample provides reference for you to learn the Ascend AI Software Stack and cannot be used for commercial purposes.

This README file provides only guidance for running the sample in command line (CLI) mode.

Thanks for the sample contribution of Tsinghua University.

portrait_picture Sample

Function: replaces the background of an input portrait with a new one by using the PortraitNet model.
Input: a JPG portrait and a JPG background image
Output: the portrait with a new background

Prerequisites

Check whether the following requirements are met. If not, perform operations according to the remarks. If the CANN version is upgraded, check whether the third-party dependencies need to be reinstalled. (The third-party dependencies for 5.0.4 and later versions are different from those for earlier versions.)

Item	Requirement	Remarks
CANN version	≥ 5.0.4	Install the CANN by referring to Installation in the About Ascend Samples Repository. If the CANN version is earlier than the required version, switch to the samples repository specific to the CANN version. See Release Notes.
Hardware	Atlas200DK/Atlas300 (Ai1s)	Currently, the Atlas 200 DK and Atlas 300 have passed the test. For details about the product description, see Hardware Platform. For other products, adaptation may be required.
Third-party dependency	opencv,python-acllite	Select required dependencies. See Third-Party Dependency Installation Guide (Python Sample).

Sample Preparation

1. Obtain the source package.

   You can download the source code in either of the following ways:

   • Command line (The download takes a long time, but the procedure is simple.)

     # In the development environment, run the following commands as a non-root user to download the source repository:   
     cd ${HOME}     
     git clone https://github.com/Ascend/samples.git
     

     Note: To switch to another tag (for example, v0.5.0), run the following command:

     git checkout v0.5.0
     

   • Compressed package (The download takes a short time, but the procedure is complex.)
     Note: If you want to download the code of another version, switch the branch of the samples repository according to the prerequisites.

      # 1. Click Clone or Download in the upper right corner of the samples repository and click Download ZIP.   
      # 2. Upload the .zip package to the home directory of a common user in the development environment, for example, ${HOME}/ascend-samples-master.zip.    
      # 3. In the development environment, run the following commands to unzip the package:    
      cd ${HOME}    
      unzip ascend-samples-master.zip
     

2. Obtain the required model.

Model	Description	How to Obtain
Portrait	TensorFlow-based portrait segmentation model.	Download the original model and the corresponding .cfg file by referring to README.md.

# To facilitate download, the commands for downloading the original model and converting the model are provided here. You can directly copy and run the commands. You can also refer to the above table to download the model from ModelZoo and manually convert it.    
cd $HOME/samples/python/contrib/portrait_picture/model
wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/003_Atc_Models/AE/ATC%20Model/PortraitNet%20/portrait.pb   
wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/003_Atc_Models/AE/ATC%20Model/PortraitNet%20/insert_op.cfg
atc --model=./portrait.pb  --insert_op_conf=./insert_op.cfg  --output="./portrait" --output_type=FP32 --input_shape="Inputs/x_input:1,224,224,3" --framework=3 --soc_version=Ascend310

• Note: If you are converting models on the 310B chip, please set the value of "soc_version" parameter as "Ascend310B1" before running the atc command.

3. Obtain the test image required by the sample.

   # Run the following commands to go to the data folder of the sample and download the corresponding test image:
   cd $HOME/samples/python/contrib/portrait_picture/data
   wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/models/Portrait/background.jpg
   wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/models/Portrait/ori.jpg
   

Sample Running

Note: If the development environment and operating environment are set up on the same server, skip step 1 and go to step 2 directly.

1. Run the following commands to upload the portrait_picture directory in the development environment to any directory in the operating environment, for example, /home/HwHiAiUser, and log in to the operating environment (host) as the running user (HwHiAiUser):

   # In the following information, xxx.xxx.xxx.xxx is the IP address of the operating environment. The IP address of Atlas 200 DK is 192.168.1.2 when it is connected over the USB port, and that of Atlas 300 (AI1s) is the corresponding public IP address.
   scp -r $HOME/samples/python/contrib/[email protected]:/home/HwHiAiUser
   ssh [email protected]
   

2. Run the executable file.

   • If step 1 has been performed, please run the following command

     cd ${HOME}/portrait_picture/src   
     python3 main.py
     

   • If not, please run the following command

     cd $HOME/samples/python/contrib/portrait_picture/src
     python3 main.py
     

Result Viewing

After the execution is complete, find the result JPG image in the out/result directory.

Common Errors

For details about how to rectify the errors, see Troubleshooting. If an error is not included in Wiki, submit an issue to the samples repository.