tensorrtx icon indicating copy to clipboard operation
tensorrtx copied to clipboard

retinface seems to fail to serialize on Jetson Orin AGX 64 GB

Open whitewater69 opened this issue 3 months ago • 1 comments

Env

  • GPU : Jetson Orin AGX 64 Gb dev liot
  • OS : Ubuntu 20.04
  • Cuda version : 11.4
  • TensorRT version : 8,5,2,1

About this repo

  • Branch : Master
  • Model : retinface

Your problem

When running sudo ./retina_r50 -s to serialize the model, the script seems to fail.

[03/08/2024-08:29:04] [W] [TRT] The implicit batch dimension mode has been deprecated. Please create the network with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag whenever possible. Loading weights: ../retinaface.wts Your platform support int8: 1 Building engine, please wait for a while... reading calib cache: r50_int8calib.table [03/08/2024-08:29:11] [W] [TRT] Unknown embedded device detected. Using 59662MiB as the allocation cap for memory on embedded devices. [03/08/2024-08:29:14] [E] [TRT] 1: Unexpected exception _Map_base::at Build engine successfully! retina_r50: /home/slacroix/retinaface/tensorrtx/retinaface/retina_r50.cpp:251: void APIToModel(unsigned int, nvinfer1::IHostMemory**): Assertion `engine != nullptr' failed. Aborted

Thank you for your help

whitewater69 avatar Mar 08 '24 13:03 whitewater69