I am trying to produce TensorRT engine for my Tensorflow model.
First step, TF model is converted to ONNX.
During conversion I have three errors.
2020-08-03 21:47:18,431 - ERROR - Tensorflow op [CTCGreedyDecoder: CTCGreedyDecoder] is not supported 2020-08-03 21:47:18,431 - ERROR - Tensorflow op [d_predictions: SparseToDense] is not supported 2020-08-03 21:47:18,431 - ERROR - Unsupported ops: Counter({'CTCGreedyDecoder': 1, 'SparseToDense': 1})
What should I do with these errors?
The whole conversion process is as follows.
I used opset12
python -m tf2onnx.convert --graphdef numplate_recg.pb --opset 12 --output numplate_recg.onnx --inputs input:0 --outputs d_predictions:0
2020-08-03 21:47:15,796 - WARNING - From /home/itc/venv/lib/python3.6/site-packages/tf2onnx/verbose_logging.py:76: The name tf.logging.set_verbosity is deprecated. Please use tf.compat.v1.logging.set_verbosity instead.
2020-08-03 21:47:18,198 - INFO - Using tensorflow=1.14.0, onnx=1.7.0, tf2onnx=1.6.3/d4abc8
2020-08-03 21:47:18,198 - INFO - Using opset <onnx, 12>
2020-08-03 21:47:18,431 - ERROR - Tensorflow op [CTCGreedyDecoder: CTCGreedyDecoder] is not supported
2020-08-03 21:47:18,431 - ERROR - Tensorflow op [d_predictions: SparseToDense] is not supported
2020-08-03 21:47:18,431 - ERROR - Unsupported ops: Counter({'CTCGreedyDecoder': 1, 'SparseToDense': 1})
2020-08-03 21:47:18,439 - INFO - Optimizing ONNX model
2020-08-03 21:47:18,638 - INFO - After optimization: Cast -4 (9->5), Const -17 (62->45), Identity -3 (3->0), Reshape +2 (3->5), Shape -1 (4->3), Squeeze -2 (4->2), Transpose -37 (41->4), Unsqueeze -3 (3->0)
2020-08-03 21:47:18,643 - INFO -
2020-08-03 21:47:18,643 - INFO - Successfully converted TensorFlow model numplate_recg.pb to ONNX
2020-08-03 21:47:18,658 - INFO - ONNX model is saved at numplate_recg.onnx