Add benchmarking mode to loitering-detection and smart-parking #793
Conversation
Signed-off-by: Sairam Pillai <[email protected]>
sairampillai
requested review from
ArokiEdgard,
ajagadi1,
rrajore,
tjanczak,
vagheshp and
xwu2intel
as code owners
Signed-off-by: Sairam Pillai <[email protected]>
...te/metro-vision-ai-app-recipe/loitering-detection/src/dlstreamer-pipeline-server/config.json
Outdated
Show resolved
Hide resolved
...ai-suite/metro-vision-ai-app-recipe/smart-parking/src/dlstreamer-pipeline-server/config.json
Show resolved
Hide resolved
...ai-suite/metro-vision-ai-app-recipe/smart-parking/src/dlstreamer-pipeline-server/config.json
Show resolved
Hide resolved
|
Please setup meeting with me to describe your benchmark design: How it is working and how your KPIs are calculated. This is critical to ensure the right information is delivered. |
Signed-off-by: Sairam Pillai <[email protected]>
Signed-off-by: Sairam Pillai <[email protected]>
| "name": "smart_parking_benchmarking", | ||
| "source": "gstreamer", | ||
| "queue_maxsize": 50, | ||
| "pipeline": "multifilesrc location=/home/pipeline-server/videos/new_video_1.mp4 loop=true ! parsebin ! vah264dec ! vapostproc ! video/x-raw(memory:VAMemory) ! gvadetect name=detection ! queue ! gvaclassify name=classification ! queue ! gvapython module=/home/pipeline-server/models/colorcls2/process class=Process function=process_frame ! queue ! gvawatermark ! videoconvertscale ! video/x-raw,format=I420 ! gvafpscounter ! appsink sync=false async=false", |
There was a problem hiding this comment.
please replace "parsebin ! vah264dec ! vapostproc ! video/x-raw(memory:VAMemory)" with "decodebin3"
There was a problem hiding this comment.
please also remove "videoconvertscale ! video/x-raw,format=I420" - we should not need to copy output buffer to CPU-side memory
| "detection-properties": { | ||
| "model": "/home/pipeline-server/models/intel/pedestrian-and-vehicle-detector-adas-0001/FP16-INT8/pedestrian-and-vehicle-detector-adas-0001.xml", | ||
| "device": "GPU", | ||
| "inference-interval": 5, |
There was a problem hiding this comment.
Why do we set different inference-interval for GPU (5) and CPU (3)?
| "inference-region": 0, | ||
| "batch-size": 1, | ||
| "nireq": 4, | ||
| "pre-process-backend": "va-surface-sharing", |
There was a problem hiding this comment.
please use "va" or NPU;
"va-surface-sharing" is not supported, and there is internal fallback to "va" inside DLStreamer - let's show end customers the actual backend used
|
This is a dup of #875 ? |
Description
Add Benchmarking mode for Metro AI Vision apps loitering-detection and smart-parking.
How Has This Been Tested?
Tested using end to end loitering-detection and smart-parking pipeline.
Checklist: