Segmentation fault when do OTA with segmentation model

[Icancode]

Could you guys help me confirm if OTA update supports segmentation model?

• Hardware Platform (GPU)
• DeepStream Version: nvcr.io/nvidia/deepstream:6.3-gc-triton-devel

• Issue Type( questions, bugs)
I modified deepstream-test5 example a bit like this:

• add pre-process plugin before nvinfer
• using model segmentation instead in the config-file properties of nvinfer plugin

Problem: Segmentation fault when switching segmentation model with deepstream example

• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
Could you download my deepstream-test6 example to reproduce the issue?
Run app:

./deepstream-test5-app -c configs/test5_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt -p 0 -o configs/test5_ota_override_config.txt 

Then modifiy the ota config

vim configs/test5_ota_override_config.txt

deepstream-test6.zip (77.2 KB)

[fanzh]

sorry for the late reply. could you use gdb to get the crash stack?

[Icancode]

here it is:

[fanzh]

please refer to section 8 of test5 README. there are some limitation for OTA functionality.

[Icancode]

I didn’t use any other models, I just switched the same model but got this error

[fanzh]

how did you operate exactly? let me have a try.

[Icancode]

just download my example app, I just add nvpreprocess before nvinfer then run app like README and do OTA file modification

[fanzh]

nvinfer plugin is opensource. I can reproduce this issue. it is related to if (oldParams.numDetectedClasses){} part code in DsNvInferImpl::initNewInferModelParams. when I removed this part code. the app run fine. will continue to check.

[Icancode]

let me know if you have final solution, thank you so much

[fanzh]

numDetectedClasses is only for detector and Instance Segmentation model. here is the final solution.
In function DsNvInferImpl::initNewInferModelParams of /opt/nvidia/deepstream/deepstream/sources/gst-plugins/gst-nvinfer/gstnvinfer_impl.cpp, please modify
if (oldParams.numDetectedClasses) {
to
if ((oldParams.networkType == NvDsInferNetworkType_Detector
|| oldParams.networkType == NvDsInferNetworkType_InstanceSegmentation)
&& oldParams.numDetectedClasses) {
then rebuild according to readme, then replace /opt/nvidia/deepstream/deepstream/lib/gst-plugins/libnvdsgst_infer.so with the new so.

[user64451]

thank you so much, I will give a try and respond to you soon

[fanzh]

Sorry for the late reply, Is this still an DeepStream issue to support? Thanks!