TensorFlow Implementation (load mode & expose out param)

[carlos-aguero]

@michaelgruner will pass pointer for download to @GFallasRR - Nov 26

[carlos-aguero]

Make R2I classes: Frame, Prediction, Loader, Factory, Model, Engine, Parameters (Expose parameters: output name, input names).

Acceptance Criteria:

1. Run TensorFlow working with GoogleNet

[carlos-aguero]

model running now, but not valid output... Michael converted module to NCSDK and output is acceptable now, Michael will check for a new model and validate!

Making code at .c for testing and understanding and after running and validating will create required classes

[carlos-aguero]

Model runs at C, init image normalization was causing problems, people is detected now. Next expose with classes

[carlos-aguero]

Classes (Ready) factory, model, loader Classes (TODO) Frame, Prediction, Engine, Parameters (Expose parameters: output name, input names).

[carlos-aguero]

Unit tests for all classes (except frame and prediction)

[carlos-aguero]

• Model loaded

• Input loaded and converted to Tensor

• Creates prediction that is an output tensor

• Will be delivered without tests

• Tests will be another task @ #21

• [ ] Implement engine predict function and input/output parameters

[carlos-aguero]

• [ ] input/output parameters not ready
• [ ] not merge at develop
• test not ready (other task) #21

Tested with valgrind, no leaks reported

[carlos-aguero]

tests are:

1. loader DO IT
2. frame DO IT prediction On Hold engine On Hold model On Hold parameters On Hold