- A grayscale image of shape (256, 64, 1)
-
Apply Conv2D with
32 filters,kernel size = (3,3),padding='same' -
Apply Batch Normalization
-
Apply ReLU Activation
-
Apply MaxPooling2D
(pool size = (2,2)) -
Apply Conv2D with
64 filters,kernel size = (3,3),padding='same' -
Apply Batch Normalization
-
Apply ReLU Activation
-
Apply MaxPooling2D
(pool size = (2,2)) -
Apply Dropout
(rate=0.2) -
Apply Conv2D with
128 filters,kernel size = (3,3),padding='same' -
Apply Batch Normalization
-
Apply ReLU Activation
-
Apply MaxPooling2D
(pool size = (2,1)) -
Apply Dropout
(rate=0.2)
- Reshape the output to
(64, 1024), making it a sequence of 64 steps with 1024 features each.
- Apply a Dense Layer with
64 units - Apply a Bidirectional LSTM layer with
512 units - Apply another Bidirectional LSTM layer with
512 units
- Apply a Dense Layer with
30 units(number of output classes) - Apply Softmax Activation to generate probability scores
- A sequence output of shape (64, 30) representing
64 time stepswith30 possible classes