DeepStream output file metadata and metrics

[michele.salvatore]

• Hardware Platform (Jetson / GPU) NVIDIA L4-12Q
• DeepStream Version 7.1
• TensorRT Version 10.3.0.26
• NVIDIA GPU Driver Version (valid for GPU only) 550.54.14

Hello,

I’m currently using DeepStream 7.1 with a deepstream_app configuration file (deepstream_appv8_config.txt).

[application]
enable-perf-measurement=1
perf-measurement-interval-sec=5

[tiled-display]
enable=1
rows=1
columns=1
width=640
height=640
gpu-id=0
nvbuf-memory-type=0

[source0]
enable=1
type=4
uri=rtsp:/url:port/mystream
num-sources=1
gpu-id=0
cudadec-memtype=0
latency=200

[sink0]
enable=1
type=1
sync=1
gpu-id=0
nvbuf-memory-type=0

[sink2]
enable=1
type=4
gpu-id=0
rtsp-port=8554
enc-type=1
#udp-port=5400
bitrate=8000000
profile=0
iframeinterval=10
nvbuf-memory-type=0

[osd]
enable=1
gpu-id=0
border-width=1
text-size=15
text-color=1;1;1;1;
text-bg-color=0.3;0.3;0.3;1
font=Serif
show-clock=0
clock-x-offset=800
clock-y-offset=820
clock-text-size=12
clock-color=1;0;0;0
nvbuf-memory-type=0

[streammux]
gpu-id=0
live-source=1
batch-size=1
batched-push-timeout=40000
width=640
height=640
enable-padding=0
nvbuf-memory-type=0

[primary-gie]
enable=1
gpu-id=0
gie-unique-id=1
nvbuf-memory-type=0
config-file=config_infer_primary_yolov8.txt
bbox-border-color0=1;0;0;1
bbox-border-color1=0;1;1;1
bbox-border-color2=0;0;1;1
bbox-border-color3=0;1;0;1

The pipeline is working correctly and outputs an RTSP stream as expected.

However, I would like to enable the output of detection metadata and metrics, such as:

• Bounding box information
• Object coordinates
• Class IDs, confidence scores
• Any other inference-related data

My goals are:

1. To generate a file (e.g., JSON, CSV, etc.) during runtime containing these detection results.
2. Nice to Have: publish these metadata outputs to a message broker, such as MQTT or Kafka.

Could someone please clarify:

• What blocks (e.g., [message-converter], [message-broker], [tracker], etc.) and parameters I need to add to the config file to enable this?
• Whether it’s possible to produce both a local file and publish metadata to MQTT/Kafka at the same time using deepstream-app?
• Do I need to write a custom probe function or can this be configured entirely via the config file?

Thanks in advance for your help!

Best regards,

[yuweiw]

About the msgbroker and msgconvert, you can refer to our sample sources\apps\sample_apps\deepstream-test5\configs.
About the “produce in a local file” issue, you can just set the gie-kitti-output-dir=streamscl in the config file.

[michele.salvatore]

Hello,
I followed the guide that you provided and I am now able to produce messages on kafka topics.
It is not clear how to handle the types of output posted on the topic.

I am sending you an example of the output that I am getting:

kafka:

{
“version”: “4.0”,
“id”: “150”,
“@timestamp”: “2025-06-26T08:23:05.368Z”,
“sensorId”: “0”,
“objects”: [
“18446744073709551615|0.871582|38.8585|636.922|598.878|Defected”
]
}

I would like to be able to extract all the metadata included in the files created in the folder of this parameter “gie-kitti-output-dir” and publish them to kafka

example of output in gie-kitti-output-dir:

Defected 0.0 0 0.0 49.589722 41.170227 640.000000 603.489441 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.823895

I currently composed the file config like this:

[application]
enable-perf-measurement=1
perf-measurement-interval-sec=5
gie-kitti-output-dir=/opt/nvidia/deepstream/deepstream-7.1/yolo/logs

[tiled-display]
enable=1
rows=1
columns=1
width=640
height=640
gpu-id=0
nvbuf-memory-type=0

[source0]
enable=1
#type=3
#uri=udp://127.0.0.1:5000
type=4
uri=rtsp://url:port/mystream
num-sources=1
gpu-id=0
cudadec-memtype=0
latency=200

[sink0]
enable=1
type=1
sync=1
gpu-id=0
nvbuf-memory-type=0

[sink2]
enable=1
type=4
gpu-id=0
rtsp-port=8554
enc-type=1
#udp-port=5400
bitrate=8000000
profile=0
iframeinterval=10
nvbuf-memory-type=0

[sink3]
enable=1
type=6
msg-broker-proto-lib=/opt/nvidia/deepstream/deepstream/lib/libnvds_kafka_proto.so
msg-broker-conn-str=url;port
topic=detections
msg-conv-config=config_msg_kafka.txt
msg-broker-comp-id=1
msg-conv-comp-id=1
msg-conv-payload-type=1
new-api=1
msg-conv-msg2p-new-api=1

config_msg_kafka.txt:

[msg-conv-msg2p]
enable=1
payload-type=1
msg2p-include-objects=1
msg2p-include-bbox=1
bbox-min-width=1
bbox-min-height=1
frame-interval=1
msg2p-include-extended=1

Could you please help me?

Thanks @yuweiw

[yuweiw]

You need to modify our source code to send all the data. You can refer to the https://forums.developer.nvidia.com/t/how-to-add-custom-data-to-analytics-or-send-data-via-kafka-inside-deep-stream/247944.