Spresense SDK - Tensorflow : Flatbuffers missing

Arco

Hello I am trying to recreate Tensorflow demo called :"micro_speech" located : "tensorflow/lite/micro/examples/micro_speech/"

However during a build I am getting : "fatal error: flatbuffers/flatbuffers.h: No such file or directory"

I really don't know how to solve this issue?
Is there anyway how to import this library ?

KamilTomaszewski

Hi @Arco

The flatbuffers library is downloaded during the build process. Please check if you successfully downloaded the flatbuffers library while building. You can check this in the logs after running make command.

After a successful download, the library will be here: externals/tensorflow/tensorflow-<SHA1>/tensorflow/lite/micro/tools/make/downloads/flatbuffers

Best Regards,
Kamil Tomaszewski

Arco

Hello @KamilTomaszewski

Thank you very much for your help.

Yes the flatbuffers library was missing, unfortunately I was not able to run make command so I manually launched flatbuffers_download.sh script which successfully downloaded this library. However problem still persists

Thanks, Juraj

Arco

Here is messages :

In file included from /Users/arcius/spresenseenv/usr/arm-none-eabi/include/c++/7.3.1/type_traits:35:0,
                 from /Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/core/api/flatbuffer_conversions.h:24,
                 from /Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/micro/micro_mutable_op_resolver.h:23,
                 from /Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/micro/all_ops_resolver.h:19,
                 from VoiceRecognition.cxx:3:
/Users/arcius/spresenseenv/usr/arm-none-eabi/include/c++/7.3.1/bits/c++0x_warning.h:32:2: error: #error This file requires compiler and library support for the ISO C++ 2011 standard. This support must be enabled with the -std=c++11 or -std=gnu++11 compiler options.
 #error This file requires compiler and library support \
  ^~~~~
In file included from /Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/core/api/flatbuffer_conversions.h:28:0,
                 from /Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/micro/micro_mutable_op_resolver.h:23,
                 from /Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/micro/all_ops_resolver.h:19,
                 from VoiceRecognition.cxx:3:
/Users/arcius/C++/Sony-Spresense/spresense/sdk/../externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/schema/schema_generated.h:21:10: fatal error: flatbuffers/flatbuffers.h: No such file or directory
 #include "flatbuffers/flatbuffers.h"

KamilTomaszewski

Hi @Arco

What OS are you using?

Arco

@KamilTomaszewski

BigSur ( Macbook M1 )

KamilTomaszewski

@Arco

I think the problem may be that TensorFlow doesn't support building on Mac.

Could you please try to build this example on any target: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/micro/examples/hello_world
Follow the instructions from the README.

Arco

@KamilTomaszewski

I can see ( from build logs ) that tensorflow sources are build successfully. The output builds can be found in

/spresense/externals/tensorflow/tensorflow-372e7eef27e03adabceb4c7ca41d366776573a73/tensorflow/lite/micro/tools/make/gen/spresense_cortex-m4_micro

however when use this includes in my project :

#include "tensorflow/lite/micro/all_ops_resolver.h"
#include "tensorflow/lite/micro/micro_error_reporter.h"
#include "tensorflow/lite/micro/micro_interpreter.h"
#include "tensorflow/lite/schema/schema_generated.h"
#include "tensorflow/lite/version.h"

I am getting error mentioned above.

In SDK Config I set 'Spresense SDK' -> 'Externals' -> 'Tensorflow lite micro' to 'Y'.

Also in Makefile of my application ( command ) I set

# C++ compiler flags
CXXFLAGS = -std=c++11

To avoid those

error: #error This file requires compiler and library support for the ISO C++ 2011 standard. This support must be enabled with the -std=c++11 or -std=gnu++11 compiler options.
#error This file requires compiler and library support \

Is there anything that I forgot ? ( I don't think that problem is with TensorFlow and Mac ).

Thanks

KamilTomaszewski

@Arco

Since you are trying to recreate Tensorflow demo called :"micro_speech", can you try to run these commands:

spresense/sdk$ make distclean
spresense/sdk$ ./tools/config.py examples/tf_example_micro_speech
spresense/sdk$ make

This will allow you to build and run "micro_speech" demo on Spresense. Flash .spk, open nuttx shell and run this command:

nsh> tf_example

I would like to first check if you can successfully run it. Please let me know.

Arco

@KamilTomaszewski

tf_example_micro_speech was successfully build -> flashed -> launched ( working correctly ) on my Spresense.

KamilTomaszewski

@Arco Thank you for your confirmation.

The problem is that the compiler doesn't have the path given where the flatbuffers library is. Even if you add it, you will encounter other errors. This is because there are two build systems. The first is NuttX which has its own implementations of standard libraries. The second is TensorFlow which uses the standard libraries from the compiler. When you try to build something with TensorFlow on NuttX then you have compatibility problems with standard libraries.

If you want to create a new application then you must include it in the TensorFlow system build. Then, in the NuttX application, call functions like in this example:
https://github.com/sonydevworld/spresense/blob/master/examples/tf_example

It is currently not possible to call TensorFlow functions directly from the NuttX application. This is only temporary because we are working to make it possible. This is the first SDK release that has allowed the TensorFlow examples to run on Spresense and unfortunately has some limitations.

I am sorry for the inconvenience.
Best Regards,
Kamil Tomaszewski

Arco

@KamilTomaszewski Thank for your help I own you one

Do you think that another solutions like CircuitPython or MicroEJ would be better for using TensorFlow ?

Thank you again

KamilTomaszewski

@Arco You're welcome.

Unfortunately, TensorFlow is currently only supported on the NuttX SDK.

Arco

Hello @KamilTomaszewski

after a while I came back to this issue.
I tried to create new project inside of /spresense/examples (same as tf_examples). I was following whole structure of tf_examples from Makefile, Kconfig to creating configuration files in /spresense/sdk/examples/.

I have launched ( hlasove_povely is name of the project )

$ tools/config.py hlasove_povely

and after that

$ make

But I am getting this output :

make[6]: *** No rule to make target `context'.  Stop.
make[5]: *** [hlasove_povely_context] Error 2
make[4]: *** [../examples/_context] Error 2
make[3]: *** [/Users/arcius/C++/Sony-Spresense/spresense/sdk/apps/.._context] Error 2
make[2]: *** [context_serialize] Error 2
make[1]: *** [context] Error 2
make: *** [all] Error 2

Am I something missing thanks

Arco

Right now after calling :

tools/config.py hlasove_povely

the response is :

make[6]: *** No rule to make target `preconfig'.  Stop.
make[5]: *** [Kconfig] Error 2
make[4]: *** [../examples/_preconfig] Error 2
make[3]: *** [preconfig] Error 2
make[2]: *** [Kconfig] Error 2
make[1]: *** [apps_preconfig] Error 2
make: *** [olddefconfig] Error 2
Post process failed. 512
Try 'make distclean' first.

(I tried 'make distclean' but didn't help)

Thanks

KamilTomaszewski

Hi @Arco

Do you include in your Makefile: include $(APPDIR)/Application.mk

The error could be because it is missing.

Arco

@KamilTomaszewski
Yes I do.

Makefile :

-include $(TOPDIR)/Make.defs
-include $(SDKDIR)/Make.defs

PROGNAME = $(CONFIG_EXAMPLES_HLASOVE_POVELY_PROGNAME)
PRIORITY = $(CONFIG_EXAMPLES_HLASOVE_POVELY_PRIORITY)
STACKSIZE = $(CONFIG_EXAMPLES_HLASOVE_POVELY_STACKSIZE)
MODULE = $(CONFIG_EXAMPLES_HLASOVE_POVELY)

MAINSRC = hlasove_povely_main.c

include $(APPDIR)/Application.mk

Make.defs :

ifneq ($(CONFIG_EXAMPLES_HLASOVE_POVELY),y)
CONFIGURED_APPS += hlasove_povely
endif

Kconfig :

config EXAMPLES_HLASOVE_POVELY
	tristate "Tensorflow Example app"
	default n
	---help---
		Enable the NuttX command line application for executing example of
		Tensorflow lite micro.

if EXAMPLES_HLASOVE_POVELY

config EXAMPLES_HLASOVE_POVELY_PROGNAME
	string "Program name"
	default "hlasove_povely"
	---help---
		This is the name of the program that will be use when the NSH ELF
		program is installed.

config EXAMPLES_HLASOVE_POVELY_PRIORITY
	int "hlasove_povely task priority"
	default 100

config EXAMPLES_HLASOVE_POVELY_STACKSIZE
	int "hlasove_povely stack size"
	default 32768

endif

KamilTomaszewski

Hi @Arco

Could you try to create a new example using this command:

sdk$ ./tools/mkcmd.py hlasove_povely

Arco

@KamilTomaszewski

Response is :

hlasove_povely already exists

KamilTomaszewski

@Arco

Delete or move your old example out of sdk and create a new one using this command.

Upcoming maintenance

Spresense SDK - Tensorflow : Flatbuffers missing