yolov5 exported onnx model unable to convert to TensorRT

[lchunleo]

Hi

i tried a simple example of getting a pretrained yolov5 model (yolov5s.pt) and followed the steps to export as yolov5s.onnx

However, when i tried the convert function, it returns the following errors:

03/02/2022-12:56:20] [TRT] [E] 4: [network.cpp::validate::2963] Error Code 4: Internal Error (Layer Slice_4 failed validation)
Traceback (most recent call last):
File "convert.py", line 72, in
convert_onnx_to_trt(opt.weights,
File "convert.py", line 25, in convert_onnx_to_trt
convert(np.random.randint(0, 255, image_shape).astype("float"))

py", line 200, in _build_engine
trt_engine_file.write(engine.serialize())
AttributeError: 'NoneType' object has no attribute 'serialize'

Am using : python convert.py --weights yolov5s.onnx --img-size 720 1080 instead of yolov5.engine indicate in the documentation

[shivamswarnkar]

Hi @lchunleo can you share your TensorRT and pycuda versions? Also, link to that pretrained Yolov5 weight (if it's publicly available)?

[shivamswarnkar]

staled, closing the issue. Feel free to reopen it if needed

[naga24]

@lchunleo Try using the trtexec utility from Tensorrt to convert onnx to trt it worked for me