Output shape in Deepstream

• Hardware Platform (Jetson / GPU) dGPU
• DeepStream Version 5.1
• JetPack Version (valid for Jetson only)
• TensorRT Version 7.2.1.6
• NVIDIA GPU Driver Version (valid for GPU only) 460.73.01

Hi, I have a question about output shape of video or image from Deepstream. I have a working pipelines in Deepstream, for video: uridecodebin->streammux->pgie->nvvideoconverter->videotemplate->nvosd->nvvideoconverter->capsfilter->encoder->codeparser->container->filesink and for image: filesrc->jpegparse->nvv412decoder->streammux->pgie->nvvideoconverter->videotemaplate->nvosd->nvvideoconverter->capsfilter->jpegenc->jpegparse->filesink. Everything works fine when I set width and height for streammux to 1280x720 or 1920x1080 but when I set squared shape, i.e. 1280x1280 or vertical orientation like 720x1280 the output is totally broken. I know that width has to by multiply by 8 and height by 4, but has it be also typical shape for video? I can’t find any information about it so I’m not sure if there is a bug in my implementation or it is normal for deepstream. And of course when I set output to 720x1280 the input is also 720x1280. It is a frame form input video:

It is an example of output:


And when I run deepstream with input 720x1280 but set in streammuxer 1280x720 the output is weird but without artefacts:

Can you try enable–padding?

https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_plugin_gst-nvstreammux.html#gst-properties

I tried but output looks the same

@kesong any update?

Can you share anything to let me reproduce it in my side?

Sorry for late response, I partly found the reason of the issue. The problem showed up when I converted image from RGBA to RGB and then I returned to RGBA and pushed image downstream. The image looked ok but in the final video it looked like the image above. So for now I create copy of frame, convert it to RGB and after processing I push edited frame to first three channel of original image. I have no idea why this issue exist only with vertical images and also I can edit alpha channel and it still works so it is not a problem with values in alpha channel. So I found solution but if you have any idea why it works this way I will really appreciate your explanation.
Thanks

Glad to know you found the solution.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.