Attempt at a keras implementation of TinyYoloModel

keras_tinyYoloModel_summary.txt

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         (None, 416, 416, 3)       0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 416, 416, 16)      432       
_________________________________________________________________
batch_normalization_1 (Batch (None, 416, 416, 16)      64        
_________________________________________________________________
leaky_re_lu_1 (LeakyReLU)    (None, 416, 416, 16)      0         
_________________________________________________________________
max_pooling_2d_1 (MaxPooling (None, 208, 208, 16)      0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 208, 208, 32)      4608      
_________________________________________________________________
batch_normalization_2 (Batch (None, 208, 208, 32)      128       
_________________________________________________________________
leaky_re_lu_2 (LeakyReLU)    (None, 208, 208, 32)      0         
_________________________________________________________________
max_pooling_2d_2 (MaxPooling (None, 104, 104, 32)      0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 104, 104, 64)      18432     
_________________________________________________________________
batch_normalization_3 (Batch (None, 104, 104, 64)      256       
_________________________________________________________________
leaky_re_lu_3 (LeakyReLU)    (None, 104, 104, 64)      0         
_________________________________________________________________
max_pooling_2d_3 (MaxPooling (None, 52, 52, 64)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 52, 52, 128)       73728     
_________________________________________________________________
batch_normalization_4 (Batch (None, 52, 52, 128)       512       
_________________________________________________________________
leaky_re_lu_4 (LeakyReLU)    (None, 52, 52, 128)       0         
_________________________________________________________________
max_pooling_2d_4 (MaxPooling (None, 26, 26, 128)       0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 26, 26, 256)       294912    
_________________________________________________________________
batch_normalization_5 (Batch (None, 26, 26, 256)       1024      
_________________________________________________________________
leaky_re_lu_5 (LeakyReLU)    (None, 26, 26, 256)       0         
_________________________________________________________________
max_pooling_2d_5 (MaxPooling (None, 13, 13, 256)       0         
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 13, 13, 512)       1179648   
_________________________________________________________________
batch_normalization_6 (Batch (None, 13, 13, 512)       2048      
_________________________________________________________________
leaky_re_lu_6 (LeakyReLU)    (None, 13, 13, 512)       0         
_________________________________________________________________
max_pooling_2d_6 (MaxPooling (None, 13, 13, 512)       0         
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 13, 13, 1024)      4718592   
_________________________________________________________________
batch_normalization_7 (Batch (None, 13, 13, 1024)      4096      
_________________________________________________________________
leaky_re_lu_7 (LeakyReLU)    (None, 13, 13, 1024)      0         
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 13, 13, 512)       4718592   
_________________________________________________________________
batch_normalization_8 (Batch (None, 13, 13, 512)       2048      
_________________________________________________________________
leaky_re_lu_8 (LeakyReLU)    (None, 13, 13, 512)       0         
_________________________________________________________________
conv2d_9 (Conv2D)            (None, 13, 13, 6)         3072      
_________________________________________________________________
flatten_1 (Flatten)          (None, 1014)              0         
_________________________________________________________________
fc (Dense)                   (None, 5)                 5075      
=================================================================
Total params: 11,027,267
Trainable params: 11,022,179
Non-trainable params: 5,088
_____________________________

tinyyolomodel.py

#%% YOLO in Keras from scratch?
np.random.seed(123)
from keras.layers import Input, Dense, Activation, ZeroPadding2D, BatchNormalization, Flatten, Conv2D
from keras.layers import AveragePooling2D, MaxPooling2D, Dropout, GlobalMaxPooling2D, GlobalAveragePooling2D
from keras.layers.advanced_activations import LeakyReLU
from keras.models import load_model, Model
from keras.preprocessing.image import ImageDataGenerator

def TinyYoloModel(input_shape=(416,416,3)):
    """
    Implementation of the Tiny YOLOv2 Model in Keras.

    Arguments:
    input_shape -- shape of the images of the dataset

    Returns:
    model -- a Model() instance in Keras
    """

    # Define the input placeholder as a tensor with shape input_shape. Think of this as your input image!
    X_input = Input(shape=input_shape)

    #https://github.com/xslittlegrass/CarND-Vehicle-Detection/blob/master/vehicle%20detection.ipynb
    # Layer 1-8
    for i in range(1,9):
        if i==1:
            X = X_input

        # 2D Convolution
        if i != 8:
            X = Conv2D(2**(3+i), kernel_size=(3,3), strides=(1,1), padding='same', name='conv2d_'+str(i), use_bias=False)(X)
        elif i == 8:
            X = Conv2D(512, kernel_size=(3,3), strides=(1,1), padding='same', name='conv2d_'+str(i), use_bias=False)(X)

        # Batch Normalization
        X = BatchNormalization(name='batch_normalization_'+str(i))(X)

        # Leaky ReLU
        X = LeakyReLU(alpha=0.1, name='leaky_re_lu_'+str(i))(X)

        # MaxPooling2D
        if i <= 5:
            X = MaxPooling2D(pool_size=(2,2), name='max_pooling_2d_'+str(i))(X)
        if i == 6:
            X = MaxPooling2D(pool_size=(1,1), name='max_pooling_2d_'+str(i))(X)

    # Layer 9
    '''
    Parameters:
    bbox_count -- number of bounding boxes you want each YOLO grid to output per grid cell (e.g. 5 in YOLOv2)
    bbox_info -- Confidence score for a bbox, and bbox dimension's x-position, y-position, width, and height (i.e. fixed at 5)
    classes -- number of classes you want to predict (e.g. building, tree, road, etc)
    
    So if predicting on:
    1 bounding box per grid, 1 class, filters = 1*(5+1) = 6
    5 bounding boxes per grid, 20 classes, filters = 5*(5+20) = 125 (e.g. PASCAL VOC )
    filters = (no. of bounding boxes you want model to predict) * (no. of classes + 5) 
    '''
    bbox_info = 5  #fixed at 5!!
    bbox_count, classes = (1, 1)
    last_filters = bbox_count * (bbox_info + classes)
    X = Conv2D(filters=last_filters, kernel_size=(1,1), strides=(1,1), padding='same', name='conv2d_9', use_bias=False)(X)

    # FLATTEN X (means convert it to a vector) + FULLYCONNECTED
    X = Flatten()(X)
    X = Dense(5, activation='relu', name='fc')(X)

    # Create model. This creates your Keras model instance, you'll use this instance to train/test the model.
    model = Model(inputs = X_input, outputs = X, name='TinyYoloModel')

    return model

tinyYoloModel = TinyYoloModel()
tinyYoloModel.summary()