UniAD evaluation

[senaerdogan]

Hello,

I saw that at the end of the inference_app of uniad-trt we get some visualizations. In the original UniAD, there is a command for evaluating the model. Could you explain how can we get these kind of evaluation results after running inference_app?

Thanks.

[chengzhe-xu]

Hi @senaerdogan , thank you for being interested in the UniAD-TRT project!

I think you can try to replace the PyTorch inference code in the original evaluation scripts with the TRT inference code. The general code pattern should looks like:
From: data preparation&preprocess -> PyTorch inference -> output postprocess
To: data preparation&preprocess -> set tensor address & move data from CPU to GPU -> enqueueV3 (or enqueueV2) -> move data from GPU to CPU -> output postprocess
You can refer to the inference app for more details for each steps

Hope those information is helpful. If you need further help or clarification, please let me know. Thank you!