Closed JosipNizetic closed 3 years ago
Hi,
I would like to add a bit of documentation, but right now is in an early state.
You should be able to load the .pb model with the new changes of Cppflow2, you can take a look to the example on how to load a pb model and run it in the examples folder.
Ok thank you.
Tried that but had a bit trouble. Will try to dig deeper.
Best regards, Josip
What did you tried and what are you trying to implement?
Josip's models have multiple output tensors, which is not supported by v2 yet.
On Mon, Oct 12, 2020, 1:40 PM Sergio Izquierdo notifications@github.com wrote:
What did you tried and what are you trying to implement?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/serizba/cppflow/issues/55#issuecomment-707256777, or unsubscribe https://github.com/notifications/unsubscribe-auth/AALDTHQRNRFQUZG6GUU5OBDSKM5PZANCNFSM4SMJ375Q .
Thank you very much for clarification.
Josip
@JosipNizetic One temporary workaround for this is using the v1 version in master branch and PR #27. However, the long term goal is to have SavedModel support in v2.
Hi @JosipNizetic,
I've created a function in model to work with multiple input/outpus, you can check an example on how it works on the multi_input_output example.
You can now call your model with a list of tuples ("input_name", input_tensor) and a list of the names of the outputs. You can inspect the names of the model with the CLI tool provided by tensorflow:
saved_model_cli show --dir path/to/model --all
Then you can call your cppflow model like:
model({{"serving_default_input_1:0", input_1}, {"serving_default_input_2:0", input_2}}, {"StatefulPartitionedCall:0", "StatefulPartitionedCall:1"});
Note that you can specify the index of the operations after the colon :
, as it is shown by saved_model_cli
.
Hope it helps!
@serizba Maybe using std::unordered_map
as both inputs and outputs is easier to work with, instead of a vector of tuple, although there is greater overhead of constructing a hash table.
Thank you very much. Been out few days. Will try by tomorrow and get back..
Thank you once more..
Hi @JosipNizetic,
Did you manage to make it work?
Sorry for this vey long Delay. I thought i answered it. Yes everything is working fine..
Thank you very much.
Hello,
I am sorry for bothering you but is there some more help documentation that you would suggest to use your code with following model. 033_Hand_Detection_and_Tracking
and especially palm detection model that can be find here.. saved_model_palm_detection_builtin
I would like to implement TF .pb version and see speed difference in regards to already imeplented tflite.
Thank you very much, Josip