Issue with nvdsanalytics detecting objects inside ROI when they are not

Hardware Platform:
GPU NVIDIA L40S
• DeepStream Version:
7.1
• TensorRT Version:
Aligned with DeepStream 7.1 recommendations
• Issue Type:
Questions
• Requirement Details:
Hello,

I am experiencing an issue with the nvdsanalytics plugin in DeepStream.
Objects are being reported as inside the ROI even when they are clearly outside of it.

I have attached an example image that shows the problem: the bounding boxes of some objects are detected as being within the ROI, although visually they are not.
Here an example:

Could this be related to how the ROI is defined (e.g., polygon coordinates or scaling)? Or is it possible that the plugin is using a different reference point?

Has anyone else experienced this behavior, and is there a recommended way to fix or work around it?

Here my config file:

[source0]
enable=1
type=4
uri=<rtsp>
num-sources=1
latency=200
drop-frame-interval=2
rtsp-reconnect-interval-sec=30
select-rtp-protocol=4
camera-fps-n=15
camera-fps-d=1

... <N sources>

[sourceN]
enable=1
type=4
uri=<rtsp>
num-sources=1
latency=200
drop-frame-interval=2
rtsp-reconnect-interval-sec=30
select-rtp-protocol=4
camera-fps-n=15
camera-fps-d=1


[primary-gie]
enable=1
batch-size=20
gie-unique-id=1
labelfile-path=labels.txt
nvbuf-memory-type=0
config-file=config_infer_primary.txt

[sink0]
enable=1
type=4
sync=0
codec=1
nvbuf-memory-type=0
bitrate=4000000
iframeinterval=10
rtsp-port=8554
udp-port=5555
profile=0
udp-buffer-size=1000000
qos=0

[sink1]
type=6
enable=1
...

[streammux]
live-source=1
batch-size=20
batched-push-timeout=4000
width=960
height=540
enable-padding=1
nvbuf-memory-type=0
attach-sys-ts-as-ntp=0

[osd]
enable=1
border-width=1
text-size=1
text-color=1;1;1;1;
text-bg-color=0.7;0.7;0.7;0.8
font=Arial
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

[tiled-display]
enable=1
rows=6
columns=6
width=960
height=540

[tracker]
enable=1
tracker-width=320
tracker-height=320
ll-lib-file=../libs/libnvds_nvmultiobjecttracker.so
ll-config-file=config_tracker_NvDCF_accuracy.yml
display-tracking-id=1

[nvds-analytics]
enable=1
config-file=config_nvdsanalytics.txt

My nvdsanalytics config file:

[property]
enable=1
osd-mode=1
display-font-size=2
config-width=960
config-height=540

[roi-filtering-stream-0]
enable=1
inverse-roi=0
class-id=-1
roi-zone_1=549;534;566;327;599;276;659;235;707;217;750;207;785;198;793;207;742;235;703;264;683;342;694;444;700;534;

[roi-filtering-stream-N]
enable=1
inverse-roi=0
class-id=-1
roi-zone_N=...

Thanks in advance for your support.

Can you provide a simple sample to reproduce the issue? Including the video, model, and the configurations.

Would it be possible to share such materials privately? Unfortunately, I cannot post the video, model, or configuration files publicly as they contain sensitive data belonging to our client.

You may send the forum message

@Fiona.Chen I have privately sent you a message; may I ask if you have received it?

Sorry for the late reply! With the video and cfg file you shared. I am not able to reproduce the accuracy issue. Since there is no postprocess code of the yolo model you shared, I used the native model resnet18_trafficcamnet instead. In the log 0917.txt (46.5 KB), Objs in ROI zone_0 is always 0. Here is the complete code. deepstream-nvdsanalytics-test-new.zip (46.3 KB)
if you can still reproduce this issue, please provide the complete code project. Thanks!