Hi,
I see from the previous questions, TIDL does not support recursive layers. However, a recursive layer could be converted into a fixed sequential network with shared weights.
We have converted our model to a sequential model, however, it takes multiple inputs to the network and also has a for loop to iterate over the inputs.
Does TIDL support multiple inputs to the network?
The following things have been tried:
1) TIDL Import tool was utilized. The error was following :
[libprotobuf FATAL /workspace/d3_sdk/ti-processor-sdk-rtos-j721e-evm-07_02_00_06//protobuf-3.5.1/src/google/protobuf/repeated_field.h:1522] CHECK failed: (index) < (current_size_):
terminate called after throwing an instance of 'google::protobuf::FatalException'
what(): CHECK failed: (index) < (current_size_):
Another question here: There were examples to give input images as calibration for networks. If there is a different format, how does the input list of data supposed to work?
2) After that onnx runtime was utilized to see which operations were not supported and also to see if it could be run in a heterogeneous way.
However, the logs don't capture anything, as in error logs. I wanted to know if this is because of multiple inputs to the network? or could you suggest a way to debug this?
We get an error -> 20706 segmentation fault (core dumped)
Thank You