DeepStream custom classifier parser data is not showing up on bounding box or metadata for object

I have written a custom classifier parser for a custom model based on the example given in the custom parser examples on github. https://github.com/Kwull/deepstream-4.0.1/tree/master/sources/libs/nvdsinfer_customparser

I know the custom function is running because I am able to see output that I have printed to the console that matches the classifications that I am expecting, but the data is lost somewhere. It doesn’t show up on the bounding boxes that are drawn on the output image and it doesn’t show up in the classifier metadata for each object. Any ideas?

Which model are you using for the classifier model? Can you try deepstream-test2 or deepstream-app -t with source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8_jetson.txt under $DS_TOP/samples/configs/deepstream-app

I have the same problem

@bcao I’m using a custom omnx model for the classifier and I tried both things you mentioned. The test apps work in their default state, but as soon as I introduce the model and custom classifier parser, they crash with an NVDSINFER_CUSTOM_LIB_FAILED message. In my custom python script, the model and custom classifier parser appear to run correctly (outside of not displaying the metadata from the classifier) and I am able to see that the data got to the NvDsInferContextImpl::fillClassificationOutput function because I printed out the attributes that were returned by the custom classifier in that function by recompiling the library with some print statements. I can try to keep following the function calls up as far as I can, but I feel like there is some piece of information that I am missing that would resolve it more cleanly.

@bcao I got to run correctly, but it only works synchronously. When I try to run asynchronously, it doesn’t attach the metadata. Is there any other config settings that would allow this to work asynchronously.

Try enable classifier-async-mode, refer https://docs.nvidia.com/metropolis/deepstream/plugin-manual/index.html#page/DeepStream%20Plugins%20Development%20Guide/deepstream_plugin_details.html#wwpID0E0YFB0HA