def tiny_yolo3_ultralite_mobilenetv3small_body(inputs, num_anchors, num_classes, alpha=1.0): '''Create Tiny YOLO_v3 Ultra-Lite MobileNetV3Small model CNN body in keras.''' mobilenetv3small = MobileNetV3Small(input_tensor=inputs, weights='imagenet', include_top=False, alpha=alpha) print('backbone layers number: {}'.format(len(mobilenetv3small.layers))) # input: 416 x 416 x 3 # activation_31(layer 165, final feature map): 13 x 13 x (576*alpha) # expanded_conv_10/Add(layer 162, end of block10): 13 x 13 x (96*alpha) # activation_22(layer 117, middle in block8) : 26 x 26 x (288*alpha) # expanded_conv_7/Add(layer 114, end of block7) : 26 x 26 x (48*alpha) # activation_7(layer 38, middle in block3) : 52 x 52 x (96*alpha) # expanded_conv_2/Add(layer 35, end of block2): 52 x 52 x (24*alpha) # NOTE: activation layer name may different for TF1.x/2.x, so we # use index to fetch layer # f1 :13 x 13 x (576*alpha) f1 = mobilenetv3small.layers[165].output # f2: 26 x 26 x (288*alpha) f2 = mobilenetv3small.layers[117].output f1_channel_num = int(576*alpha) f2_channel_num = int(288*alpha) #f1_channel_num = 1024 #f2_channel_num = 512 y1, y2 = tiny_yolo3_ultralite_predictions((f1, f2), (f1_channel_num, f2_channel_num), num_anchors, num_classes) return Model(inputs, [y1,y2])
def tiny_yolo3_ultralite_ghostnet_body(inputs, num_anchors, num_classes): '''Create Tiny YOLO_v3 Ultra-Lite GhostNet model CNN body in keras.''' ghostnet = GhostNet(input_tensor=inputs, weights='imagenet', include_top=False) print('backbone layers number: {}'.format(len(ghostnet.layers))) # input: 416 x 416 x 3 # blocks_9_0_relu(layer 291, final feature map): 13 x 13 x 960 # blocks_8_3_add(layer 288, end of block8): 13 x 13 x 160 # blocks_7_0_ghost1_concat(layer 203, middle in block7) : 26 x 26 x 672 # blocks_6_4_add(layer 196, end of block6) : 26 x 26 x 112 # blocks_5_0_ghost1_concat(layer 101, middle in block5) : 52 x 52 x 240 # blocks_4_0_add(layer 94, end of block4): 52 x 52 x 40 # NOTE: activation layer name may different for TF1.x/2.x, so we # use index to fetch layer # f1: 13 x 13 x 960 f1 = ghostnet.layers[291].output # f2: 26 x 26 x 672 f2 = ghostnet.layers[203].output # f3: 52 x 52 x 240 #f3 = ghostnet.layers[101].output f1_channel_num = 960 f2_channel_num = 672 #f3_channel_num = 240 y1, y2 = tiny_yolo3_ultralite_predictions( (f1, f2), (f1_channel_num, f2_channel_num), num_anchors, num_classes) return Model(inputs, [y1, y2])
def tiny_yolo3_ultralite_peleenet_body(inputs, num_anchors, num_classes): '''Create Tiny YOLO_v3 Ultra-Lite PeleeNet model CNN body in keras.''' peleenet = PeleeNet(input_tensor=inputs, weights='imagenet', include_top=False) print('backbone layers number: {}'.format(len(peleenet.layers))) # input: 416 x 416 x 3 # re_lu_338(layer 365, final feature map): 13 x 13 x 704 # re_lu_307(layer 265, end of stride 16) : 26 x 26 x 512 # re_lu_266(layer 133, end of stride 8) : 52 x 52 x 256 # NOTE: activation layer name may different for TF1.x/2.x, so we # use index to fetch layer # f1: 13 x 13 x 704 f1 = peleenet.layers[365].output # f2: 26 x 26 x 512 f2 = peleenet.layers[265].output f1_channel_num = 704 f2_channel_num = 512 y1, y2 = tiny_yolo3_ultralite_predictions( (f1, f2), (f1_channel_num, f2_channel_num), num_anchors, num_classes) return Model(inputs, [y1, y2])
def tiny_yolo3_ultralite_mobilenetv2_body(inputs, num_anchors, num_classes, alpha=1.0): '''Create Tiny YOLO_v3 Ultra-Lite MobileNetV2 model CNN body in keras.''' mobilenetv2 = MobileNetV2(input_tensor=inputs, weights='imagenet', include_top=False, alpha=alpha) print('backbone layers number: {}'.format(len(mobilenetv2.layers))) # input: 416 x 416 x 3 # out_relu: 13 x 13 x 1280 # block_13_expand_relu: 26 x 26 x (576*alpha) # block_6_expand_relu: 52 x 52 x (192*alpha) # f1 :13 x 13 x 1280 f1 = mobilenetv2.get_layer('out_relu').output # f2: 26 x 26 x (576*alpha) f2 = mobilenetv2.get_layer('block_13_expand_relu').output f1_channel_num = int(1280 * alpha) f2_channel_num = int(576 * alpha) #f1_channel_num = 1024 #f2_channel_num = 512 y1, y2 = tiny_yolo3_ultralite_predictions( (f1, f2), (f1_channel_num, f2_channel_num), num_anchors, num_classes) return Model(inputs, [y1, y2])