Environment Variables (EVs)

The table below lists the environment variables (EVs) that can be used as inputs for the container running the inferencing pipeline.

GST profile EVsCommon EVs

This list of EVs specifically supports the GST profile DLStreamer workloads.

Variable	Description	Values
AGGREGATE	aggregate branches of the gstreamer pipeline, if any, at the end of the pipeline	"", "gvametaaggregate name=aggregate", "aggregate branch. ! queue"
BARCODE_RECLASSIFY_INTERVAL	time interval in seconds for barcode classification	Ex: 5
CLASSIFICATION_OPTIONS	extra classification pipeline instruction parameters	"", "reclassify-interval=1 batch-size=1 nireq=4 gpu-throughput-streams=4"
CPU_SW_DECODER	force to use software decoder for gst-launch decoding video frames in CPU	"force-sw-decoders=1"
DECODE	decoding element instructions for gst-launch to use	Ex: "decode bin force-sw-decoders=1"
DETECTION_OPTIONS	extra object detection pipeline instruction parameters	"", "gpu-throughput-streams=4 nireq=4 batch-size=1"
GST_DEBUG	for running pipeline in gst debugging mode	0, 1
GST_PIPELINE_LAUNCH	for launching gst pipeline script file path and name	Ex: "/home/pipeline-server/framework-pipelines/yolov5_pipeline/yolov5s.sh"
LOG_LEVEL	log level to be set when running gst pipeline	ERROR, INFO, WARNING, and more
OCR_RECLASSIFY_INTERVAL	time interval in seconds for OCR classification	Ex: 5
PARALLEL_PIPELINE	run pipeline in parallel using the tee branch	"", "! tee name=branch ! queue"
PARALLEL_AGGRAGATE	aggregate parallel pipeline results together, paired use with PARALLEL_PIPELINE	"", "! gvametaaggregate name=aggregate ! gvametaconvert name=metaconvert add-empty-results=true ! gvametapublish name=destination file-format=2 file-path=/tmp/results/r$cid_count.jsonl ! fpsdisplaysink video-sink=fakesink sync=true --verbose branch. ! queue !"
VA_SURFACE	use video analytics surface from the shared memory if applicable	"", "! "video/x-raw(memory
BATCH_SIZE	number of frames batched together for a single inference to be used in gvadetect batch-size element	0, 1

This list of EVs is common for all profiles.

Variable	Description	Values
AUTO_SCALE_FLEX_140	allow workload to manage autoscaling	1, 0
CPU_ONLY	for overriding inference to be performed on CPU only	1, 0
DEVICE	for setting device to use for pipeline run	"GPU", "CPU", "AUTO", "MULTI:GPU,CPU"
LOW_POWER	for running pipelines using GPUs only	1, 0
OCR_DEVICE	optical character recognition device	"CPU", "GPU"
PRE_PROCESS	pre process command to add for inferencing	"pre-process-backend=vaapi-surface-sharing", "pre-process-backend=vaapi-surface-sharing pre-process-config=VAAPI_FAST_SCALE_LOAD_FACTOR=1"
PIPELINE_PROFILE	for choosing OVMS workload's pipeline profile to run	use make list-profiles to see Values
RENDER_MODE	for displaying pipeline and overlay CV metadata	1, 0
STREAM_DENSITY_MODE	for starting pipeline stream density testing	1, 0
STREAM_DENSITY_FPS	for setting stream density target fps value	Ex: 15.0
STREAM_DENSITY_INCREMENTS	for setting incrementing number of pipelines for running stream density	Ex: 1

Applying EV to Run Pipeline

EV can be applied in two ways:

1. as parameter input to run.sh script
2. in the env files

The input parameter will override the one in the env files if both are used.

EV as input parameter

Example

PIPELINE_PROFILE="object_detection" CPU_ONLY=1 RENDER_MODE=0 sudo -E ./run.sh --platform core --inputsrc rtsp://127.0.0.1:8554/camera_0

Note

Those EVs in front of run.sh like CPU_ONLY, RENDER_MODE are applied to this run only and they are also known as command line environment overrides, or environment overrides. They will override the default values in env files if any.

Editing the Env Files

EV can be configured for advanced user in configs/opencv-ovms/envs/. As an example for gst pipeline profile, there are two Env files can be configured:

    - `yolov5-cpu.env` file for running pipeline in core system
    - `yolov5-gpu.env` file for running pipeline in gpu or multi

These two files currently hold the default values.