head-count-algo/onnx2trt.py

import tensorrt as trt
import onnx
import torch

device = torch.device('cuda:0')

onnx = "yolo-crowd-n.onnx"

print(f"starting export of onnx to tensorrt engine... with TensorRT version: {trt.__version__}\n")
f = "yolov5-crowd-n.engine"
logger = trt.Logger(trt.Logger.INFO)

builder = trt.Builder(logger)
config = builder.create_builder_config()
config.max_workspace_size = 1 << 30

flag = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
network = builder.create_network(flag)
parser = trt.OnnxParser(network, logger)
if not parser.parse_from_file(onnx):
    print('ERROR: Failed to parse the ONNX file.')
    for error in range(parser.num_errors):
        print(parser.get_error(error))
    exit(1)

inputs = [network.get_input(i) for i in range(network.num_inputs)]
outputs = [network.get_output(i) for i in range(network.num_outputs)]
for inp in inputs:
    print(f'input: {inp.name}, shape: {inp.shape} {inp.dtype}')
for out in outputs:
    print(f'output: {out.name}, shape: {out.shape} {out.dtype}')

if builder.platform_has_fast_fp16:
    config.set_flag(trt.BuilderFlag.FP16)

build = builder.build_serialized_network

with build(network, config) as engine, open(f, "wb") as t:
    t.write(engine)