End-to-end deep learning model for natural gesture recognition powered by Keras.
- Copy preprocessed data to
/data
(Wed Mar 27 14:24:26 CET 2019) - Move hardcoded input size to config.json
(NUM_FRAMES, HEIGHT, WIDTH, ...)
inbuild_model()
(Wed Mar 27 14:24:26 CET 2019) - Create
deepswipe_model.py
(Wed Mar 27 14:24:26 CET 2019) - Create
deepswipe_data_loader.py
- Create
deepswipe_trainer.py
- Train model
- Create a data pipeline from raw data to processed data (get rid of files <10 frames, take last 10 frames)
- Refactor `config.trainer.crop : ['first','last','random']
- Gather new training data
- Installation
- Getting Started
- Running The Demo Project
- Comet.ml Integration
- Template Details
- Acknowledgements
Clone the repository
git clone https://github.com/Rainymood/keras-deepswipe
cd keras-deepswipe
Set-up the virtual environment
virtualenv --system-site-packages -p python3 venv
Note that /venv
is included in .gitignore
. Now enter the virutal environment
. venv/bin/activate
Install dependencies
pip install -r requirements.txt
In order to get started we have to:
- Define a data loader class.
- Define a model class that inherits from BaseModel.
- Define a trainer class that inherits.
- Define a configuration file with the parameters needed in an experiment.
- Run the model using:
python main.py -c [path to configuration file]
A simple model for the mnist dataset is available to test. To run the demo project:
- Start the training using:
python main.py -c configs/simple_mnist_config.json
- Start Tensorboard visualization using:
tensorboard --logdir=experiments/<YYYY-MM-DD>/simple_mnist/logs
Note: One has to fill in the appropriate date.
python main_deepswipe.py -c configs/deepswipe_config.json
And then for the visualisation you can run
tensorboard --logdir=experiments/<YYYY-MM-DD>/deepswipe/logs
Support for Comet.ml is integrated. This allows you to see all your hyper-params, metrics, graphs, dependencies and more including real-time metrics.
Add your API key in the configuration file:
For example: "comet_api_key": "your key here"
├── main.py - here's an example of main that is responsible for the whole pipeline.
│
│
├── base - this folder contains the abstract classes of the project components
│ ├── base_data_loader.py - this file contains the abstract class of the data loader.
│ ├── base_model.py - this file contains the abstract class of the model.
│ └── base_train.py - this file contains the abstract class of the trainer.
│
│
├── model - this folder contains the models of your project.
│ └── simple_mnist_model.py
│ └── i3d_inception.py
│ └── deepswipe_model.py
│
│
├── trainer - this folder contains the trainers of your project.
│ └── simple_mnist_trainer.py
│ └── deepswipe_trainer.py
│
|
├── data_loader - this folder contains the data loaders of your project.
│ └── simple_mnist_data_loader.py
│ └── deepswipe_data_loader.py
│
│
├── configs - this folder contains the experiment and model configs of your project.
│ └── simple_mnist_config.json
│ └── deepswipe_config.json
│
│
├── data - this folder might contain the datasets of your project.
│
│
└── utils - this folder contains any utils you need.
├── config.py - util functions for parsing the config files.
├── dirs.py - util functions for creating directories.
└── utils.py - util functions for parsing arguments.
You need to:
- Create a model class that inherits from BaseModel.
- Override the build_model function which defines your model.
- Call build_model function from the constructor.
You need to:
- Create a trainer class that inherits from BaseTrainer.
- Override the train function which defines the training logic.
Note: To add functionalities after each training epoch such as saving checkpoints or logs for tensorboard using Keras callbacks:
- Declare a callbacks array in your constructor.
- Define an init_callbacks function to populate your callbacks array and call it in your constructor.
- Pass the callbacks array to the fit function on the model object.
Note: You can use fit_generator instead of fit to support generating new batches of data instead of loading the whole dataset at one time.
You need to:
- Create a data loader class that inherits from BaseDataLoader.
- Override the get_train_data() and the get_test_data() functions to return your train and test dataset splits.
Note: You can also define a different logic where the data loader class has a function get_next_batch if you want the data reader to read batches from your dataset each time.
You need to define a .json file that contains your experiment and model configurations such as the experiment name, the batch size, and the number of epochs.
We can now load models without having to explicitly create an instance of each class. Look at:
- from_config.py: this can load any config file set up to point to the right modules/classes to import
- Look at configs/simple_mnist_from_config.json to get an idea of how this works from the config. Run it with:
python from_config.py -c configs/simple_mnist_from_config.json
- See conv_mnist_from_config.json (and the additional data_loader/model) to see how easy it is to run a different experiment with just a different config file:
python from_config.py -c configs/conv_mnist_from_config.json
This project builds off the Keras Project Template.