- model training
- model testing
- model evaluation
- file explorer
- bulk image upload
- screenshot script (go)
- image crawler/scrapper (go)
- graph generation
- dark/light mode
-
Clone the repository.
git clone https://github.com/bethropolis/myia.git
-
Install the dependencies by running the following command:
pip install -r requirements.txt
-
Setup project:
python setup.py
-
create a virtual environment (optional):
python -m venv myenv
activate the virtual environment
# windows myenv\Scripts\activate # linux / mac source myenv/bin/activate
-
run the app:
python app.py
open your browser and go to http://localhost:5000
-
Upload images to the app
- upload images to the
training directory
- head to
http://localhost:5000/directory?path=training/train
and upload your images
- upload images to the
-
label the images
-
Open the
train
page (http://localhost:5000/train
) -
label the images either as good or bad by clicking the
thumbs up
orthumbs down
button. -
the app can only generate binary classification models so you can only label the images as
good
orbad
thumbs up
forgood
which could represent classificationA
thumbs down
forbad
which could represent classificationB
currently the app only supports two labels
good
andbad
-
-
build the model
-
To build the model, head to the
home
page (http://localhost:5000/
) and click thebuild model
button -
In the next page you will have to input:
No of epochs
- the number of times the model will be trained (default is 15)No of layers
- the number of layers the model will have (default is 3)Model name
- the name of the model (default ismyia_image_classifier
) -
click the
build model
button to start building the model
The model will be saved in the
model/image_model
directory as akeras
modelNote: The higher the number of epochs the longer it will take to build the model
-
-
test the model - To test the model, open the
test
page (http://localhost:5000/test
) and upload an image to test the model with or test with images in thetest
directory (http://localhost:5000/directory?path=training/test
) -
evaluate the model
-
To evaluate the model, open the
evaluate
page (http://localhost:5000/evaluate
) and upload an image to evaluate the model with or evaluate with images in theevaluate
directory (http://localhost:5000/directory?path=model/evaluation
) -
The evaluation results will be saved in the
model
directory as ajson
file and a graph will be generated and saved in thestatic
directory as apng
file
-
Feel free to ping me a pull requests if you want to contribute.
This project is licensed under the MIT License.
happy coding 💜