model trained on grayscale data set of images (no_imgages , width ,height ,1) ,where no_images is 7750.
Train_set samples : 7750
Test_set samples = Validation_set samples : 624
all the images initially have 3 channels (RGB) .
Before training , first data augmentation procedure is done on first class (NORMAL) ,since in the second class (PNEUMONIA) ,there are 3 times more images than in the first class.
after augmentation both classes have almost the same number of images.
second augmentation procedure is done during training . The images are generated in grayscale ,during data generation while training .
--------- MODEL summary ----------
Model: "sequential_1"
conv2d_3 (Conv2D) (None, 64, 64, 64) 1088
activation_3 (Activation) (None, 64, 64, 64) 0
max_pooling2d_3 (MaxPooling2 (None, 64, 64, 64) 0
conv2d_4 (Conv2D) (None, 64, 64, 32) 32800
activation_4 (Activation) (None, 64, 64, 32) 0
max_pooling2d_4 (MaxPooling2 (None, 64, 64, 32) 0
conv2d_5 (Conv2D) (None, 64, 64, 32) 4128
activation_5 (Activation) (None, 64, 64, 32) 0
max_pooling2d_5 (MaxPooling2 (None, 64, 64, 32) 0
conv2d_6 (Conv2D) (None, 64, 64, 16) 2064
activation_6 (Activation) (None, 64, 64, 16) 0
max_pooling2d_6 (MaxPooling2 (None, 64, 64, 16) 0
flatten (Flatten) (None, 65536) 0
dense (Dense) (None, 256) 16777472
activation_7 (Activation) (None, 256) 0
dense_1 (Dense) (None, 128) 32896
activation_8 (Activation) (None, 128) 0
dense_2 (Dense) (None, 1) 129
Total params: 16,850,577 Trainable params: 16,850,577 Non-trainable params: 0
accuracy --- 0.9038 loss --- 0.3243