Binary or Multi Classifier to classify images by using Deep learning Architecture.
To train the model with mobilenet architectures from tensorflow.keras.applications then change the values in yaml file (parameter.yaml)
-
if MObileNet / MobileNetv2 / EfficientNetB0 to EfficientNetB7 from tensorflow.keras.applications then: sub_architecture_class as 'applications' and architecture as required architecture name (example - EfficientNetB0 or MobileNet )
-
if EfficientNetB0 to B7 from efficientnet.tfkeras then: sub_architecture_class as 'efficientnet' and architecture as 'EfficientNetB0 to EfficientNetB7' (choose any one architecture from the list)
-
if retraining is a choice then create a one diectory with in the classification-mlops and upload the base model within and copy & paste the path at architecture and 'none' at sub_architecture_class
To train multiclassifier model
-
choose number of classes and update the value of num_classes in yaml file (example - if choosen three classes then mention it as 3)
-
final_dense_activation as softmax
-
class_mode as categorical
-
loss as categorical_crossentropy
-
csv classing format will be in alphabetical order of your classes
To train binary classifier model
- choose num_classes as 2
- final_dense_activation as sigmoid
- class_mode as binary
- loss as binary_crossentropy paths -
- training path --- /data_set_name/training/train
- validation path --- /data_set_name/training/validation
- evaluation path --- /data_set_name/evalaution// Algorithms
- for normal training value is training GPU
- If utilizing the 2 gpus then hardcode the values of workers to 12 (each gpu carrys 6 workers) and choose the priority as low/medium/high.