Bad_alloc when inferring by fine-tuned pose model

[wcycqjy]

Thank you for your work. I'm trying to infer the model in c++. The platform is jetson orin nx. Everything works when I use the official yolov8s-pose model, but the bad_alloc error will appear when I use the pose model fine-tuned on personal dataset. Specifically, when my face enter the right side of the image, this error happens. I didn't change the model structure , and I see the size of official model's engine file is 26.0MB and the fine-tuned engine file is 26.1MB.

For your information, I export the model following the readme: 1. Use python script model.export(format="onnx") 2. run /usr/src/tensorrt/bin/trtexec.