2019-02-01 06:35:46.554212: I tensorflow/compiler/tf2tensorrt/convert/convert_graph.cc:193] Starting Calib Conversion
2019-02-01 06:35:46.733967: W tensorflow/compiler/tf2tensorrt/convert/convert_graph.cc:199] Construction of static int8 engine is not implemented yet!. Dynamic engine will be constructed
2019-02-01 06:35:47.901680: I tensorflow/compiler/tf2tensorrt/utils/trt_resources.cc:25] Destroying Calibration Resource
Calibrator = 0x7f0658001110
Builder = 0
Engine = 0x7f0800033a20
Logger = 0x7f0658008708
Allocator = 0x7f0658008760
Thread = 0x7f0658008740
2019-02-01 06:36:13.455979: I tensorflow/compiler/tf2tensorrt/utils/trt_resources.cc:25] Destroying Calibration Resource
Calibrator = 0x7f0648020ac0
Builder = 0
Engine = 0x7f080584ed00
Logger = 0x7f0648053a08
Allocator = 0x7f064801e7b0
Thread = 0x7f0648007740
2019-02-01 06:36:14.412340: I tensorflow/compiler/tf2tensorrt/utils/trt_resources.cc:25] Destroying Calibration Resource
Calibrator = 0x7f0698008e20
Builder = 0
Engine = 0x7f055c0c7e30
Logger = 0x7f0698008738
Allocator = 0x7f0698007bb0
Thread = 0x7f0698008ed0
pure virtual method called
terminate called without an active exception
Aborted (core dumped)
python3 tensorrt.py --frozen_graph=frozen_inference_graph.pb -bs 1 -if 1-1.jpeg --int8
https://github.com/tensorflow/models/blob/master/research/tensorrt/tensorrt.py
I’m trying to convert tensorflow object detection graph(Faster RCNN) to tensorRT int 8.
(http://download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz)
Other presicions are converted but int8 is not working.
The error appears when I call this method “calib_graph_to_infer_graph” in “import tensorflow.contrib.tensorrt as trt”.
How can I fix this?
I’m using tf-nightly-gpu (tf-nightly-gpu 1.13.0.dev20190130).
cuda version is 10 (Cuda compilation tools, release 10.0, V10.0.130)
Ubuntu 16.04
TensorRT 5.0.2
python 3.5