Hi,
I ran SSD with tensorflow without tensorrt,it ran fine on my jetson tx2. But when I tried using tensorrt, it is crashing.
Below are the parameters I am passing:
trt_graph = trt.create_inference_graph(
input_graph_def=frozen_graph,
outputs=output_names,
max_batch_size=2,
max_workspace_size_bytes=1 << 20,
precision_mode=‘FP32’,
minimum_segment_size=2
Below are the errors:
018-08-17 16:53:36.586931: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2390] Segment @scope ‘Postprocessor/BatchMultiClassNonMaxSuppression/PadOrClipBoxList/’, converted to graph
2018-08-17 16:53:36.656382: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2390] Segment @scope ‘Postprocessor/BatchMultiClassNonMaxSuppression/PadOrClipBoxList/’, converted to graph
2018-08-17 16:53:36.726445: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2390] Segment @scope ‘Postprocessor/BatchMultiClassNonMaxSuppression/PadOrClipBoxList/’, converted to graph
2018-08-17 16:53:36.796145: I tensorflow/contrib/tensorrt/convert/convert_nodes.cc:2390] Segment @scope ‘Postprocessor/BatchMultiClassNonMaxSuppression/PadOrClipBoxList/’, converted to graph
2018-08-17 16:53:36.865306: W tensorflow/contrib/tensorrt/convert/convert_graph.cc:694] Can’t determine the device, constructing an allocator at device 0
2018-08-17 16:53:36.865852: E tensorflow/contrib/tensorrt/log/trt_logger.cc:38] DefaultLogger Tensor OutputPH_0cannot be both input and output
2018-08-17 16:53:36.865904: E tensorflow/contrib/tensorrt/log/trt_logger.cc:38] DefaultLogger Network must have at least one output
2018-08-17 16:53:36.865954: W tensorflow/contrib/tensorrt/convert/convert_graph.cc:825] Engine creation for segment 0, composed of 2 nodes failed: Internal: Failed to build TensorRT engine. Skipping…
2018-08-17 16:53:36.865985: W tensorflow/contrib/tensorrt/convert/convert_graph.cc:694] Can’t determine the device, constructing an allocator at device 0
2018-08-17 16:53:36.866303: E tensorflow/contrib/tensorrt/log/trt_logger.cc:38] DefaultLogger Tensor OutputPH_0cannot be both input and output
2018-08-17 16:53:36.866364: E tensorflow/contrib/tensorrt/log/trt_logger.cc:38] DefaultLogger Network must have at least one output
2018-08-17 16:53:36.866401: W tensorflow/contrib/tensorrt/convert/convert_graph.cc:825] Engine creation for segment 1, composed of 2 nodes failed: Internal: Failed to build TensorRT engine. Skipping…
2018-08-17 16:53:36.866429: W tensorflow/contrib/tensorrt/convert/convert_graph.cc:694] Can’t determine the device, constructing an allocator at device 0
2018-08-17 16:53:36.866731: E tensorflow/contrib/tensorrt/log/trt_logger.cc:38] DefaultLogger Parameter check failed at: …/builder/Network.cpp::addScale::104, condition: inputTensor.getDimensions().nbDims >= 3
Segmentation fault (core dumped)
Thanks.