Gst-nvinfer plugins for additional network types - super resolution & denoising

[ketronium]

Hi, is there any prospect of deepstream coming packaged with additional nvinfer network type inference options beyond classification, detection, segmentation & instance segmentation? I’m interested in adding a super resolution model in my pipeline, which would replace the frame buffer with the output of that network. Happy to write my own plugins, just wondering if I’d be doing that at the same time NVIDIA is, which I’d like to avoid.

related: Looking for a simple deepstream pipeline (python) which does both super-resolution and detection of objects

thanks

[Fiona.Chen]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

@ketronium

is there any prospect of deepstream coming packaged with additional nvinfer network type inference options beyond classification, detection, segmentation & instance segmentation?

There is already “others” type in gst-nvinfer. You can set “network-type=100” and "output-tensor-meta=1“ with gst-nvinfer configuration file, the default postprocessing in gst-nvinfer will be skipped, you can do the postprocessing by yourself to output what you want.

@ketronium

I’m interested in adding a super resolution model in my pipeline, which would replace the frame buffer with the output of that network. Happy to write my own plugins, just wondering if I’d be doing that at the same time NVIDIA is, which I’d like to avoid.

Currently we don’t have this plan. The nvvideotemplate is recommended to do such job instead of gst-nvinfer since gst-nvinfer is a “in-place” transform plugin. GstBaseTransform (gstreamer.freedesktop.org)