Bounding box sample application location details

Please provide the following info (tick the boxes after creating this topic):
Software Version
DRIVE OS 6.0.8.1
DRIVE OS 6.0.6
DRIVE OS 6.0.5
DRIVE OS 6.0.4 (rev. 1)
DRIVE OS 6.0.4 SDK
other

Target Operating System
Linux
QNX
other

Hardware Platform
DRIVE AGX Orin Developer Kit (940-63710-0010-300)
DRIVE AGX Orin Developer Kit (940-63710-0010-200)
DRIVE AGX Orin Developer Kit (940-63710-0010-100)
DRIVE AGX Orin Developer Kit (940-63710-0010-D00)
DRIVE AGX Orin Developer Kit (940-63710-0010-C00)
DRIVE AGX Orin Developer Kit (not sure its number)
other

SDK Manager Version
1.9.3.10904
other

Host Machine Version
native Ubuntu Linux 20.04 Host installed with SDK Manager
native Ubuntu Linux 20.04 Host installed with DRIVE OS Docker Containers
native Ubuntu Linux 18.04 Host installed with DRIVE OS Docker Containers
other

  1. We want some details on the sample yolov3 model that is provided to visualize bounding boxes.
  2. Does the bounding box model render the 2D detections on the visualization that is run with the nvsipl_camera application?
  3. Does it also provide labels of the objects detected?
  4. Where is the location of this bounding box sample application with instructions?

Thank you!

Dear @chris.hoffman,
Are you asking about object detection sample based on yolov3? If so, please check DriveWorks SDK Reference: Basic Object Detector and Tracker Sample for details. For source code, please see /usr/local/driveworks/samples/src/dnn/sample_object_detector_tracker on docker.

@SivaRamaKrishnaNV

sorry, which docker are you referring to here?
Also, how to run the object detection sample directly with the camera visualization using nvsipl_camera?

Dear @chris.hoffman,
If you have used sdkmanager to flash the target, you can find the source code at usr/local/driveworks/samples/src/dnn/sample_object_detector_tracker on host.

You can not run object detection directly on nvsipl_camera. The object detector sample also works with live camera. So please check using live camera data in the sample.

May I know why you are using DRIVE OS 6.0.5 and not the latest release(6.0.8.1)?

@SivaRamaKrishnaNV

  1. I had to use the NGC docker to flash 6.0.5 as I couldn’t use the SDK. In that case, where can I find the built sample or example?

  2. Will I be able to get the sample if I just download the 6.0.5 OS using SDK manager on Ubuntu Host?

  3. We are on 6.0.5 because our camera drivers were tied to it.

  4. Is it possible to show bounding box visualization on the nvsipl_camera output? Or what are the other capabilities of nvsipl_camera? We want the bounding boxes to be shown on the simulated data injected into the Orin and visualized using nvsipl_camera

  5. What do you mean by live camera? Are you talking about a real camera lens connected to group A?

You can find the source code at /usr/local/driveworks/samples/src/dnn/sample_object_detector_tracker on docker

It just shows the camera display output from live camera or can take RAW camera data as input.

Yes.

nvsipl_camera is just based in NVMedia APIs. To detect bouding boxes, we need to use TensorRT framework to perform object detection and detect bouding box.

We have object detector sample in DW as a reference example which showcase object bounding box detection on live camera or recorder data. Note that, DW internally uses NVMedia(SIPL) APIs to capture camera data and TensorRT APIs to perform any DL related task.

@SivaRamaKrishnaNV

  1. Thank you for the clarity on all questions. Do you recommend copying the source code from Docker 6.0.5 to Ubuntu host and then building
    or
    directly copying the source code to Target Orin and building the application there?

  2. I am planning on using the command from the README as reference.

./sample_object_detector_tracker --input-type=[video|camera]
                            --video=[path/to/video]
                            --camera-type=[camera]
                            --camera-group=[a|b|c|d]
                            --camera-index=[0|1|2|3]
                            --slave=[0|1]
                            --tensorRT_model=[path/to/TensorRT/model]

My understanding is I can use the same camera/platform (f.e, "IMX728_RGGB_CPHY_x4") supported in nvsipl_camera but replace the application with sample_object_detector_tracker. Is this accurate?

Does that mean, you can not upgrade to latest release DRIVE OS 6.0.8.1? I remember there was some issues in DW DNN model in DRIVE OS 6.0.5 and got fixed later. I see a reported known issue in DRIVE OS 6.0.5 release notes(3823785). Can you quickly verify the /usr/local/driveworks/bin/sample_object_detector_tracker with recorder video as input on target. If you don’t find the sample on target, please install the Driveworks data and samples debians on target. You can find them under /drive/extra on docker.

DW samples can not be compiled on target and needs to cross compiled on host. Please use docker container to cross compile DW samples.

Correct. You are expected to use right value for --camera-type. Please verify the camera functionality with sample_camera (DriveWorks SDK Reference: Camera Sample)

We will be trying this out and will update the results. Adding a response so the ticket does not get closed.

1 Like

Dear @chris.hoffman ,
Could you share any update?
Can we close the topic?

Dear @chris.hoffman,
Can we close the topic? do you have any update?

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.