[TensorRT] ImportError: No module named 'tensorrt.parsers'; 'tensorrt' is not a package

 

아래와 같은 에러는 

ImportError: No module named 'tensorrt.parsers'; 'tensorrt' is not a package

 

TensorRT 예제에서 아래코드와 같이

tensorrt.parsers 의 uffparser를 import 하여 uff 파일을 로드하고 builder 를 생성하였을 때 나는 에러이다. 

 

 

import tensorflow as tf

import tensorrt as trt

from tensorrt.parsers import uffparser


uff_model = uff.from_tensorflow(tf_model, ["fc2/Relu"])
    
G_LOGGER = trt.infer.ConsoleLogger(trt.infer.LogSeverity.ERROR)
    
parser = uffparser.create_uff_parser()
parser.register_input("Placeholder", (1,28,28), 0)
parser.register_output("fc2/Relu")

engine = trt.utils.uff_to_trt_engine(G_LOGGER, uff_model, parser, 1, 1 << 20)
    
parser.destroy()

# get image
# ...

runtime = trt.infer.create_infer_runtime(G_LOGGER)
context = engine.create_execution_context()
    
# 10 is classification size
output = np.empty(10, dtype = np.float32)
    
# Alocate device memory
d_input = cuda.mem_alloc(1 * img.nbytes)
d_output = cuda.mem_alloc(1 * output.nbytes)
    
bindings = [int(d_input), int(d_output)]
    
    
stream = cuda.Stream()

# Transfer input data to device
cuda.memcpy_htod_async(d_input, img, stream)

# Execute model
context.enqueue(1, bindings, stream.handle, None)

# Transfer predictions back
cuda.memcpy_dtoh_async(output, d_output, stream)

# Syncronize threads
stream.synchronize()


print("Test Case: " + str(label))
print ("Prediction: " + str(np.argmax(output)))

trt.utils.write_engine_to_file("./tf_mnist.engine", engine.serialize())
    
    
context.destroy()
engine.destroy()
new_engine.destroy()
runtime.destroy()

 

https://docs.nvidia.com/deeplearning/sdk/tensorrt-archived/tensorrt_401/tensorrt-api/python_api/workflows/tf_to_tensorrt.html

Generating TensorRT Engines from TensorFlow — TensorRT 4.0 documentation

 

 

 

 

 

NVIDIA에 의하면 아래와 같다. 

 

 

The format has changed with the new python API. The documentation has more details on how this can be done.
https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#import_model_python

You would do something like this with the new API
with trt.Builder(TRT_LOGGER) as builder, builder.create_network() as network, trt.CaffeParser() as parser:

------------------------
To continue using the legacy API, you would need to `import tensorrt.legacy` instead of `import tensorrt`

 

 

출처 : https://devtalk.nvidia.com/default/topic/1042377/importerror-no-module-named-tensorrt-parsers-/

https://devtalk.nvidia.com/default/topic/1042377/importerror-no-module-named-tensorrt-parsers-/

 

 

uff 파일을 로드하고 builder 를 생성하여 engine 을 생성하기 위해서는 아래와 같이 변경한다. 

 

 

1. 아래와 같이 콘솔에서 pb 파일을 uff 파일로 변환 

$ convert-to-uff frozen_inference_graph.pb

 

2. import 부분 변경 및 builder 방식 수정 

import tensorrt as trt


...


model_file = '/data/mnist/mnist.uff'


with builder = trt.Builder(TRT_LOGGER) as builder, builder.create_network() as network, trt.UffParser() as parser:
    	parser.register_input("Placeholder", (1, 28, 28))
    	parser.register_output("fc2/Relu")
		parser.parse(model_file, network)