nanodet TensorRT inference demo

Hello everyone, here is a tensorrt inference demo for nanodet: https://github.com/linghu8812/tensorrt_inference/tree/master/project/nanodet.

First of all, when I export onnx model, I add softmax and concat layer to onnx, so the end of onnx model looks like this: In this way, this will increase the inference time of the model, but it will reduce the postprocessing time. Considering comprehensively, the total processing time has been reduced, so I choose this way to export the onnx model.

In addition, the onnxsim module has been imported when export onnx, so the model exported has been simplified.

from onnxsim import simplify

onnx_model = onnx.load(output_path)  # load onnx model
model_simp, check = simplify(onnx_model)
assert check, "Simplified ONNX model could not be validated"
onnx.save(model_simp, output_path)
print('finished exporting onnx ')

At last, the TensorRT inference result has shown below:

for more information, please refer: https://github.com/linghu8812/tensorrt_inference

Dec 08 '20 15:12 linghu8812

How to export nanodet onnx model with softmax and concat? I used nanodet_m.ckpt, and export-onnx.py from https://github.com/linghu8812/tensorrt_inference, but onnx model is still like this: