Hi, I run the centerface onnx model and find although faces are detected the bbox size are quite different between onnx model and tensorrt model.
You can see that the first image is the one I run use the given onnx model and the second image is the result I run with the converted tensorrt engine.
Anyone find similar issue and know how to solve this?
A bit late, but if it might help, I had same bug with TensorRT 7.0 (on nvidia TensorRT 20.03 and Triton Inference Server 20.03 images).
Updating to latest TensorRT (on 20.09 image) fixed the problem
Hi, I run the centerface onnx model and find although faces are detected the bbox size are quite different between onnx model and tensorrt model. You can see that the first image is the one I run use the given onnx model and the second image is the result I run with the converted tensorrt engine. Anyone find similar issue and know how to solve this?