YOLOX TRT model giving multiple bounding boxes while inferencing

AI & Data Science Deep Learning (Training & Inference) TensorRT
, ,
1

YOLOX TRT model giving multiple bounding boxes while inferencing

We trained a TRT model to run on our jetson agx board using this: Megvii-BaseDetection/YOLOX: YOLOX is a high-performance anchor-free YOLO, exceeding yolov3~v5 with MegEngine, ONNX, TensorRT, ncnn, and OpenVINO supported. Documentation: https://yolox.readthedocs.io/ (github.com)

ISSUE

When we are running the TRT converted inference model, it is giving us multiple bounded boxes with just the cpp code. With Python code (and same TRT model), the predictions are accurate.

TensorRT Version: 8.0.1
GPU Type: JETSON AGX
Jetpack Version: 4.6
CUDA Version: 10.2.300
CUDNN Version: 8.2.1.32
Python Version (if applicable): 3.6
PyTorch Version (if applicable): 1.10
yolox.cpp.txt (19.1 KB)

2

Hi,

We recommend you to please share with us the issue repro model and scripts to try from our end for better debugging.

Thank you.