tensor-guess

this project is just an helper with docker images based on tutorial from https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/index.html

##requirements

• docker !
• git clone this project in a directory

##usage

You just need to make a "classifier" directory with a directory "data" inside it with all your images For example

 [any_path]/my_own_classifier/
 [any_path]/my_own_classifier/data
 [any_path]/my_own_classifier/data/car
 [any_path]/my_own_classifier/data/moto
 [any_path]/my_own_classifier/data/bus

and then put your image on it. This "classifier" directory will have your samples but also trained classifier after execution of "train.sh".

##Train process

Just type

 ./train.sh [any_path]/my_own_classifier

And it will do anything for you !

##Guess process

Just type for a single guess

 ./guess.sh [any_path]/my_own_classifier /yourfile.jpg

To guess an entire directory

./guessDir.sh [any_path]/classifier [any_path]/srcDir [any_path]/destDir

Example of result

# ./guess.sh /synced/tensor-lib/moto-classifier/ /synced/imagesToTest/moto21.jpg
moto (score = 0.88331)
car (score = 0.11669)

##Remarks

Use absolute file path for classifier and images because for the moment my script do not support relative path (volume mounting) You can use pwd as a workaround