NVIDIA / TensorRT

NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.
https://developer.nvidia.com/tensorrt
Apache License 2.0
10.76k stars 2.13k forks source link

pytorch_quantization QAT on centerpoint #2447

Closed shuyuan-wang closed 1 year ago

shuyuan-wang commented 2 years ago

Have you guys tried pytorch_quantization on pointpillars or centerpoint? when I use pytorch_quantization on centerpoint and used QAT, the outcome is much worse than torch model, about 10% worse. PTQ is even worse, about 25% worse. Not sure if this is expected?

zerollzeng commented 2 years ago

@ttyio ^ ^

ttyio commented 1 year ago

@shuyuan-wang , we have not experiment on centerpoint yet, all the networks we have published in http://arxiv.org/abs/2004.09602 Maybe need you to fine-tune more or turn off some Q/DQ after sensitivity analysis.

ttyio commented 1 year ago

closing since no activity for more than 14 days, please reopen if you still have question, thanks!