Vehicle detection with trafficcamnet

Accelerated Computing Intelligent Video Analytics Transfer Learning Toolkit
#1

Hello,

Please, I try to do the detection of vehicles in a parking lot (to count them after), I am working with this tutorial https://ngc.nvidia.com/catalog/models/nvidia:tlt_trafficcamnet, on the other hand I noticed that the network can just detect vehicles that are far away, as shown in this picture
image_save5

I do not see where the problem is

I use Gstreamer
const char * streamModel = "uridecodebin uri=file:///home/sylia/Projets/DeepStream-Gstreamer/stream/image5.jpg "
"! nvvideoconvert "
"! m.sink_0 nvstreammux batch-size=1 width=960 height=540 name=m "
"! nvinfer config-file-path=configs/inference.txt "
"! nvdsosd name=osd "
"! nvvideoconvert "
"! jpegenc "
"! jifmux "
"! filesink location= image_out/image_save5.jpg "

ps: it was the same remark for all the images it always detects images far away
inference.txt (1.9 KB)

thank you

#3

Please share your setup with us

1 Like
#4

thanks
all the interference configuration is in inference.txt (1.9 KB) labels.txt (28 Bytes) resnet18_trafficcamnet_pruned.etlt (5.2 MB)

#5

@sylia,
Can you share the tool/script/command how to run the inference?

1 Like
#6

const char * streamModel = “uridecodebin uri = file: ///home/sylia/Projets/DeepStream-Gstreamer/stream/image5.jpg”
“! Nvvideoconvert”
"! M.sink_0 nvstreammux batch-size = 1 width = 960 height = 540 name = m "
"! Nvinfer config-file-path = configs / inference.txt "
"! Nvdsosd name = osd "
"! Nvvideoconvert "
"! Jpegenc "
"! Jifmux "
"! Filesink location = image_out / image_save5 .jpg "

#7

Can you try to run with default way mentioned in the https://ngc.nvidia.com/catalog/models/nvidia:tlt_trafficcamnet

Run deepstream-app :

deepstream-app -c deepstream_app_source1_trafficcamnet.txt
#8

Or run with https://github.com/NVIDIA-AI-IOT/deepstream_tlt_apps/