Please provide complete information as applicable to your setup. • Hardware Platform Jetson • DeepStream Version 6.2 • JetPack Version 5.1 I wanted to run the cityscapes_fan_tiny_hybrid_224.onnx model from here And I found some advice that is hard to parse. In the overview section, it shows ho…

[image] ganinduN: Anyway if you think this can’t be backported to at least 6.2 I’d say my question is answered and I’m happy to mark as resolved? Yes. Unfortunately, we are unable to provide a compatible library from 6.4 to 6.2 version.

We have a similar demo seg_app_unet.yml . You can see if that neet your needs.

Does that work with segformer model that I used? I want to detect and segment pixels belonging to the road.

It uses the following model citysemsegformer . You can refer to the configuration file of the demo I attached. They are similar.

Thanks, I looked at your example and got the model that was the basis of the qiestion to work (sort of ) But I don’t really understand why it is working (unless that is an illusion) becasuse the orignial advice on the model card is confliciting with what seems to work (e.g. network type, cluster mo…

[image] ganinduN: But I don’t really understand why it is working (unless that is an illusion) becasuse the orignial advice on the model card is confliciting with what seems to work Do you mean that the following configuration will not work, network-type=100 and cluster-mode=2? [image] ga…

Thanks for the quick answer , filtering out not working with segmentation explains the odd behaviour. Network mode 100 does not produce segment masks but it think pythonifying the c code above can help with that issue. (And does cluster mode 2 work with segmentation?) Gst-inspect nvsegvisusl does …

We want to make this work in deepstream 6.2. Can we patch that element (segvisual) so it has the overlay functionality. For example pull the deb file directly and install it?

If not I guess we can create a tee and do ‘nvsegvisual’ on one branch and use a mixer to blend the two and then convert back with a converter and present to a sink

Yes. Since we cannot support that for DeepStream 6.2, you can implement yourself, including the filter function.

Thanks, out of curiosity which DS does the native overlaying work for segvis?

Running Pre-trained Segformer - CityScapes our of the box from NGC

Accelerated Computing Intelligent Video Analytics DeepStream SDK

yuweiw April 7, 2024, 8:45am 17

Yes. Unfortunately, we are unable to provide a compatible library from 6.4 to 6.2 version.

1 Like

Topic		Replies	Views
How to use object detection and segmentation models in the same pipeline and get the outputs in the screen DeepStream SDK	13	2530	October 12, 2021
How to using my custom semantic segmentation model? DeepStream SDK segmentation , deepstream	11	222	May 16, 2025
Semantic Segmentation in DS 4 DeepStream SDK	12	2726	October 12, 2021
Nvidia Deepstream6.3 custom segmentation model inference error: Aborted (core dumped) DeepStream SDK python	10	352	July 23, 2024
Cascaded Instance Segmentation for defect analysis DeepStream SDK tensorrt , ubuntu , gstreamer , deepstream	17	290	October 30, 2024
Custom new model segmentation in DeepStream DeepStream SDK tensorflow	41	1792	February 3, 2023
Deepstream segmask sample with dashcamnet error DeepStream SDK	12	517	July 21, 2023
image segmentation DeepStream SDK	2	2196	October 12, 2021
Requires function to overlay with input image for nvsegvisual result DeepStream SDK	8	426	July 14, 2023
No Segmentation Meta Data (NvDsInferSegmentationMeta) built with custom segmentation model DeepStream SDK	7	1296	October 12, 2021

Running Pre-trained Segformer - CityScapes our of the box from NGC

Related topics