Pipeline error in video ingestion module while inputting a new video

[Smrithi]

Apart from the videos in VideoIngestion/test_videos in eis, I am trying to replace it with any other video keeping only the video ingestion module and removing the video analytics part. While doing so I am facing the following error:-

ERROR:OpenCvIngestor:85: Failed to open gstreamer pipeline: ./test_videos/TLC_10.50.130.40_0054213_Trim_1

I have also attached the detailed logs. Any insight is appreciated.

[WengWai_C_Intel]

Hi @Smrithi ,

Can you share your Video Ingestion configuration from the etcd web UI so that we can look at how you configure the video pipeline?

By the way, what is the input video file format in your setup? From the log file, it showing a file name without file extension. Are you using avi, mp4, or any other video file format?

Pls let us know, thanks! 

[Murni_J_Intel]

Hi, 

You can get familiarize with the configuration by trying to switch from the default PCB demo to the Safety Worder demo that included in the downloaded package as well. 

1. Open the ETCD Web UI.

2. Change the Video Ingestion Configuration  pipeline

"pipeline": "./test_videos/Safety_Full_Hat_and_Vest.avi"

3. Change the Video Ingestion Configuration  udfto use dummy

"udfs": [
{
"name": "dummy",
"type": "python",
"training_mode": "false",
"n_total_px": 300000,
"n_left_px": 1000,
"n_right_px": 1000
}
]

4. change the VideoAnalytics udf as below 

"udfs": [{
"name": "safety_gear.safety_classifier",
"type": "python",
"model_xml":"common/video/udfs/native/safety_gear_demo/ref/frozen_inference_graph.xml",
"model_bin":"common/video/udfs/native/safety_gear_demo/ref/frozen_inference_graph.bin",
"device": "CPU"
}]

5. CLick on the safe button. On the Visualizer you will see the safety_worker demo being run instead of the default PCB demo.  

By using the same concept of using dummy filter in the VideoIngestion, you can try to work with your video file.  

[Smrithi]

I have shared the SS of video ingestion configuration from etcd web UI. The video input format we are using is mp4 or avi. Despite mentioning the file format and file path correctly the video ingestion container is unable to successfully run. Every time we log into the csl manager dashboard its module status is shown as pending. 

[WengWai_C_Intel · Accepted solution]

Hi @Smrithi ,

If you are running the modified Video Ingestion container with CSL, have you rebuild the Video Ingestion container with the new video file included into the container image, then register the new Video Ingestion container image into SMR so CSL will pull the new image to run on your CSL node?

[Smrithi]

While trying to rebuild the container images using docker-compose build to include the new video file in the container image, I am getting the following error:- 

ERROR: Service 'ia_video_analytics' failed to build: The command '/bin/sh -c pip3.6 install -r va_requirements.txt && rm -rf va_requirements.txt' returned a non-zero code: 1

I have also attached the SS of the error for your understanding.

[WengWai_C_Intel]

Hi @Smrithi ,

Yes, there is a dependency package build issue with the pip version being used that just happened since last week. There will be a hotfix release soon to address this. For the time being, you can follow the quick workaround as  mentioned in the post below.

https://community.intel.com/t5/Intel-Edge-Software-Hub/Unable-to-download-eis-installer-from-edgesoftware-Gives-below/m-p/1231230#M346

[Smrithi]

@WengWai_C_Intel thank you for your reply. I am able to input other videos as well. But now, I do not want to run the video analytics container. But when I tried to build the docker images after commenting out the video analytics part in docker-compose.yml, I couldn't see the video in web visualizer. It was shown as disconnected. I even corrected the subscribe topics of other modules in the docker-compose.yml file.  And I also cannot remove the video analytics part from the csl_app_spec.json file . So, what am I supposed to do to bypass the video analytics container altogether?

[WengWai_C_Intel]

Hi @Smrithi ,

Glad to hear you are now able to use your own video file. 

For removing the video analytics part from the csl_app_spec.json, you have to recreating the new csl_app_spec.json by following the EII Master Node Setup in user guide document. You can remove the Video Analytics directory before running the eis_builder.py script so it will not include Video Analytics config in the docker compose file, module spec and app spec files. Then follow by module spec files update and etc as mentioned by the EII Master Node setup in user guide.

In fact, I would strongly suggest you to complete the development of all the containers without CSL environment first, so that you will not need to repeatedly running the deployment setup on CSL. In this case, you will also able to  easily running all your containers on the host system and verify all functionalities. Once all the containers being verified and completed development, then only deploy the container images to run on CSL environment. Please consider this approach to simplify your current workflow.

By the way, this is separate topic compare to the original of this post.  Could you help to create a new post if need further assist, so we can close this 'pipeline error in video ingestion' topic?

thank you!

[Smrithi]

As suggested by @Murni_J_Intel  we were successfully able to run the safety full hat test video as well. But when we try to input our own videos, the video ingestion module is unable to run and it's module status is always shown as pending.