T04:- Execution of existing model on the target device

       Execution of  DLC model on Target device        

In the last blog , we have seen how we can convert the existing  model into DLC. Now ,We will see how we can actually run the dlc file into Target Device.

First, we have run this model on host device itself by using the tool which is given in the SNPE.


 We have to use the path of container dlc path and input list path of running on the host.
  --container -> It is the dlc file path.
  --input_list -> It is the raw_list txt path. 

-> Now to show the output on the screen we have to use another command for showing it onto the     screen.

 Now we will see how we have actually run it on the Android Target .
1. First , we have to export the necessary libraries for running it on the Android device.
2. Second , we have to export the necessary executable files on the Android device.

3.Then we have made the necessary folders on the Android device using adb shell , we have made this on  Android target using documentation text . 

4.We have made the bin file on the Android target for snpe file executables.
5.And then we have made the lib file on the Android target for the snpe Target architecture .
6.After that we have used adb push command for running it on the Android Target.

Prepare assets in the Android device target.

    Now to pull the output file we have run this command.
               

 
         So now we will see how we can show our Output results on the screen.
     vivek@Ubuntu2:/opt/qcom/aistack/qairt/2.21.0.240401/examples/Models/InceptionV3/scripts$               python show_inceptionv3_classifications.py   -i /opt/qcom/aistack/qairt/2.21.0.240401/examples/Models/InceptionV3/data/target_raw_list.txt   -o /opt/qcom/aistack/qairt/2.21.0.240401/examples/Models/InceptionV3/tensorflow/output_android   -l /opt/qcom/aistack/qairt/2.21.0.240401/examples/Models/InceptionV3/tensorflow/imagenet_slim_labels.txt
Classification results
cropped/notice_sign.raw 0.130224 459 brass
cropped/plastic_cup.raw 0.989595 648 measuring cup
cropped/chairs.raw      0.380808 832 studio couch
cropped/trash_bin.raw   0.719754 413 ashcan

        

So, this is the overall results of running ML Model on Android Target which is Snapdragon 660.

Now to build the Android app for running this ML model Ondevice we have to run these commands.


After running these commands we can execute the following command to build the APK from the command line.

After running this command assembleDebug our APP will be assembled on Android Target.




So this is the result when running on Android Target.

Comments

Post a Comment

Popular posts from this blog

T02:- How to setup the Qualcomm Neural Processing SDK

Image
    Setup of Qualcomm Neural Processing SDK We will now understand how we can setup the Qualcomm Neural Processing Engine SDK in our system   -> To start  with first thing we have to understand what is required for this setup.       1 . Oracle Virtual Box  https://www.virtualbox.org/wiki/Downloads       2 . Ubuntu 20.04   https://releases.ubuntu.com/focal/      We can also work with WSL2 environment that will work on Windows 10/11.       3 . Any Machine Learning Framework         3.1 Tensorflow         3.2 Pytorch         3.3 Tensorflow lite         3.4 Onnx       4 . Android NDK       5 . Microsoft Visual C++(Resdistrubution) So now we will go to the practical aspect of the setup of the Neural Processing SDK(SNPE). 1 .We will first download the Virtual box as per the requirement. ...
Read more

T01:- Introduction to Ondevice AI

Image
                Understanding Of On-device Ai Smart stuff like phones and cars are getting smarter . They can crunch data and learn without needing the internet all the time. Qualcomm is a big part of making this happens.                                       Fig - 1.1(On-device AI) So, What is On-device AI? On-device means smart devices like phones or smartwatches can work without need of constant internet connection. Its like having a live brain inside your gadgets, making quick decisions and performing task without the need of live servers. This technology is good for things like voice assistants and image recognition , as it reduces the latency and ensures privacy by keeping data local.                                           Fig- 1.2 ( ON-...
Read more

Integration Workflow of Ai engine sdk

Image
                        Integration Workflow . Training and Inference workflow:- This diagram illustrates the difference between training a neural       network(done on server) and running inference(making predictions) on device. Training involves creating and optimizing a neural network model using large datasets, typically done off device. Inference is the process of using the trained model to make predictions or decisions based on new data which is usually done on device where the application runs. .Qualcomm AI engine Direct integration workflow:-  It details how to integrate trained deep learning(DL networks) into applications using Qualcomm AI engine SDK. It involves converting the trained model into format suitable  for execution on the Qualcomm chipsets and integrating them into applications. Now, lets connect these workflows with an explanation of the integration process. Integration WorkST...
Read more