bai-nvdsanalytics (master branch)

Boulder AI customizations on top of the NVIDIA deepstream nvdsanalytics project.

See Deepstream Documentation and Gst-nvdsanalytics Documentation for more details about the configuration and setup.

This master branch can be used to run inference on .mp4 files on both Jetson and dGPU platforms. To run on a Boulder AI DNNCam using the image sensor (baicamerasrc) as the input source, please check out the baicamerasrc branch of this repository and follow the README there.

---

Building

Run ./docker/build.sh

Running

Run ./docker/run.sh [docker options] where [docker options] can optionally specify volume mounts to mount volumes inside the docker container with -v <path>:<container path> where path represents the path outside the container and container path represents the path inside the container. Use these container paths:

• /data for the working base directory (see /data/ structure below). If not specified, this directory will be written to data-default in this working directory.

• /input for input files. Use /input/video.mp4 for the input video used in the default configuration. If not specified, the container will provide its own sample video as input.

• /output for output files, where the container will write its output. If not specified, the output will be written to the /data/output directory inside the container and viewable wherever the /data container directory is mounted outside the container. A file "overlay.mp4" in this directory will contain the output video overlay. An output log will also be saved in this directory.

For more about docker volume mounts, see the documentation on docker.com

Examples:

Run with default input video, writing output to data-default:

./docker/run.sh

Run with custom input video at /tmp/my-test-video.mp4 writing output to data-default:

./docker/run.sh -v /tmp/my-test-video.mp4:/input/video.mp4

Run with custom input video at /tmp/my-test-video.mp4 and write output to /tmp/data-mytestdir:

./docker/run.sh -v /tmp/my-test-video.mp4:/input/video.mp4 \
                -v /tmp/mytestdir-output:/output

/data directory structure

The /data/ directory structure will be created if necessary on first run and will contain:

• /data/cfg/deepstream containing deepstream configuration files. If no files are present in this directory at startup they will be created from the defaults in src/cfg-deepstream-default.

• /data/cfg/model containing model related files. The default model is currently peoplenet. This model is downloaded through the src/setup-peoplenet.sh script if it doesn't already exist.

Viewing Output

By default, the overlay output will be written to /output/overlay.mp4 which will be available outside the container as described above. To write RTSP instead, customize the deepstream configuration file to use the RTSP output sink and then view based on configuration there, typically rtsp://<ip address>:8554/ds-test

Modifying Settings

You can access the content of the data-dir (default data-default) outside the container. This will be mapped to /data inside the container. You can use this to add files to the container, modify the deepstream configuration, or update the model, then re-run as desired.