ImageSwift
Easily train an image recognition model on your own images!
Set Up
Please install in a virtual environment. The package requires a specific version of tensorflow.
$ pip install imageswiftUsage
The functions of ImageSwift are split into training on your dataset and then predicting on any images you want.
Setting Up Your Dataset
Please set up your image recognition project directory in the following manner:
├── Project Directory
├── dataset_name
│ ├── class1_name (each class directory should contain the images corresponding to that class)
│ ├── class2_name
│ ├── .
│ ├── .
│ ├── .
│ └── classN_name
├── train.py (see Training)
├── predict.py (see Predicting)
Training
Here is an example of how to use ImageSwift to train an image recognition model:
from imageswift import training
projectPath = "path\\to\\project\\directory"
dataset = "name_of_dataset_directory"
model = training.ImageModel(projectPath, dataset, epochs=50, image_size=(150, 150), batch_size=32, validation_split=0.2, lr=1e-3)
model.trainModel()The code above will save the the model and weights files to the project directory.
This examples also shows the defaults for some training parameters: epochs, desired image size, batch size, the percentage of your dataset to use for validation, and the learning rate.
Predicting
Here is an example of how to use ImageSwift to predict on images with the model you trained:
from imageswift import predicting
modelPath = 'path\\to\\model'
weightsPath = 'path\\to\\weights'
datasetPath = "path\\to\\dataset"
finalModel = predicting.TrainedModel(modelPath, weightsPath, datasetPath)
finalModel.loadAndCompile()
#Single Prediction
imagePath = 'path\\to\\image\\for\\prediction'
prediction = finalModel.predict(imagePath) # returns a string of the predicted class
print(prediction)Use the code below to predict on a set of images instead of one singular image:
#Accuracy on Sample Set (when predicting on a sample set, make sure it follows the same folder structure as the dataset as shown above)
samplesPath = 'path\\to\\sample\\set'
accuracy = finalModel.evaluate(samplesPath) # returns the accuracy of your model on the sample set
print(accuracy)
