/Tensorflow-Project-Signs

Simple, structured Tensorflow project.

Primary LanguagePythonApache License 2.0Apache-2.0

Intro

Simple convolution model implementation in Tensorflow (v1) for classification of two dimensional images of sign-language signs for numbers 0-5. The model and dataset is taken from the course "Convolutional Neural Networks". The dataset is not provided here, but may be downloaded by navigating to the notebook of the assignment on Coursera. Create a directory "datasets" in project root and put the two h5-files in this. Note that the code has been completely re-written to fit within the generic project template https://github.com/MrGemy95/Tensorflow-Project-Template.

Usage

Ensure that consistent versions of the various packages required for the code are used by generating a virtual environment and installing pip packages (python3) listed in requirements.txt:

pip3 install -r requirements.txt

Begin the training by running the main script:

python3 train_test_signs.py -c configs/signs.json

Setup live monitoring of the training history output in the summary using tensorboard:

tensorboard --logdir experiments/signs/summary

In a browser then open "localhost:6006" to see the graphs.

TODO

  • print accuracy at given intervals
  • add tests to summary
  • show the total iteration number when continuing from a checkpoint
  • use minibatch when testing