English|中文

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. For details about how to run the sample in MindStudio, see Running Image Samples in MindStudio.

3D Action Recognition Sample

Function: Use the 3DCNN model to perform classification inference on data. Input: Use the video2bin script to convert a video file to a data file. Output: confidence of 10 actions.

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.)

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://gitee.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 source network model required by the application.

   Model	Description	How to Obtain
   3DCNN	3D action recognition model. It is a 3DCNN model based on TensorFlow.	Download the model and weight files by referring to the links in README.md in the ATC_3DCNN_tensorflow_AE directory of the ModelZoo repository.

   # 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/3Dgesture_recognition/model    
   wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/003_Atc_Models/AE/ATC%20Model/3D_gesture_recognition/3d_gesture_recognition.pb
   atc --model=3d_gesture_recognition.pb --framework=3 --output=3d_gesture_recognition --soc_version=Ascend310 --input_shape="X:1,16,112,112,3" --input_format=NDHWC
   

3. Obtain the test images required by the sample.

   Run the following commands to go to the **data** folder of the sample and download the corresponding test images:
   cd $HOME/samples/python/contrib/3Dgesture_recognition/data
   wget https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/003_Atc_Models/AE/ATC%20Model/3D_gesture_recognition/testdata/test_float32_actiontype7.bin
   cd ../src
   

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 3Dgesture_recognition 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/3Dgesture_recognition HwHiAiUser@xxx.xxx.xxx.xxx:/home/HwHiAiUser
   ssh HwHiAiUser@xxx.xxx.xxx.xxx
   cd ${HOME}/3Dgesture_recognition/src    
   

2. Run the sample.

   python3.6 3Dgesture_recognition.py ../data/
   

Result Viewing

After the running is complete, the confidence of the 10 actions is printed on the screen.

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.