snpe-tensorflow-to-dlc conversion fails in tf2
Join Date: 28 May 21
Posts: 1
Posted: Mon, 2021-05-31 03:43
I am trying to convert a SSD Mobilenet saved model trained in tf2.3 to dlc using the snpe-tensorflow-to-dlc command as follows:
`snpe-tensorflow-to-dlc --input_network exports/saved_model -d "input_tensor" 1,300,300,3 --out_node StatefulPartitionedCall/map/TensorArrayUnstack/TensorListFromTensor -o ssdmobilenet_tf2_trained.dlc --show_unconsumed_nodes`
The conversion fails due to unsupported layers:
2021-05-27 05:10:42,525 - 405 - WARNING - ERROR_TF_FALLBACK_TO_ONDEMAND_EVALUATION: Unable to resolve operation output shapes in single pass. Using on-demand evaluation!
2021-05-27 05:10:42,525 - 188 - INFO - INFO_ALL_BUILDING_NETWORK:
==============================================================
Building Network
==============================================================
2021-05-27 05:10:42,536 - 193 - WARNING - WARNING_TF_OP_NOT_SUPPORTED: Operation (StatefulPartitionedCall/map/TensorArrayUnstack/TensorListFromTensor) of type (TensorListFromTensor) is not supported by converter.
2021-05-27 05:10:42,540 - 193 - WARNING - Output node StatefulPartitionedCall/map/TensorArrayUnstack/TensorListFromTensor, specified via command line, does not exist in graph.
2021-05-27 05:10:42,541 - 183 - ERROR - Conversion FAILED!
The same occurs when any other layer in the graph is given as the output node, which seems to imply that all layers are unsupported, which is not plausible.
Any help in solving this issue will be greatly appreciated.