def test_model21(self): model1 = Sequential(self.s, model_table='Simple_CNN1') model1.add(InputLayer(3, 224, 224)) model1.add(Conv2d(8, 7)) pool1 = Pooling(2) model1.add(pool1) conv1 = Conv2d(1, 7, src_layers=[pool1]) conv2 = Conv2d(1, 7, src_layers=[pool1]) model1.add(conv1) model1.add(conv2) model1.add(Concat(act='identity', src_layers=[conv1, conv2])) model1.add(Pooling(2)) model1.add(Dense(2)) model1.add(OutputLayer(act='softmax', n=2)) if self.data_dir is None: unittest.TestCase.skipTest(self, "DLPY_DATA_DIR is not set in the environment variables") caslib, path = caslibify(self.s, path=self.data_dir+'images.sashdat', task='load') self.s.table.loadtable(caslib=caslib, casout={'name': 'eee', 'replace': True}, path=path) r = model1.fit(data='eee', inputs='_image_', target='_label_', max_epochs=1) self.assertTrue(r.severity == 0) model1.deploy(self.data_dir, output_format='onnx')
def onnx_extract_concat(graph, node, layers): ''' Construct concat layer from ONNX op Parameters ---------- graph : ONNX GraphProto Specifies a GraphProto object. node : ONNX NodeProto Specifies a NodeProto object. layers : list of Layers Specifies the existing layers of a model. Returns ------- :class:`Concat` ''' previous = onnx_find_previous_compute_layer(graph, node) if not previous: src_names = [find_input_layer_name(graph)] else: src_names = [p.name for p in previous] src = [get_dlpy_layer(layers, i) for i in src_names] return Concat(name=node.name, act='identity', src_layers=src)
def test_model15(self): # test RECTIFIER activation for concat layer try: import onnx except: unittest.TestCase.skipTest(self, "onnx not found in the libraries") model1 = Sequential(self.s, model_table='Simple_CNN1') model1.add(InputLayer(3, 224, 224)) model1.add(Conv2d(8, 7)) pool1 = Pooling(2) model1.add(pool1) conv1 = Conv2d(1, 7, src_layers=[pool1]) conv2 = Conv2d(1, 7, src_layers=[pool1]) model1.add(conv1) model1.add(conv2) model1.add(Concat(act='RECTIFIER', src_layers=[conv1, conv2])) model1.add(Pooling(2)) model1.add(Dense(2)) model1.add(OutputLayer(act='softmax', n=2)) if self.data_dir is None: unittest.TestCase.skipTest( self, "DLPY_DATA_DIR is not set in the environment variables") caslib, path, tmp_caslib = caslibify(self.s, path=self.data_dir + 'images.sashdat', task='load') self.s.table.loadtable(caslib=caslib, casout={ 'name': 'eee', 'replace': True }, path=path) r = model1.fit(data='eee', inputs='_image_', target='_label_', max_epochs=1) self.assertTrue(r.severity == 0) import tempfile tmp_dir_to_dump = tempfile.gettempdir() model1.deploy(tmp_dir_to_dump, output_format='onnx') import os os.remove(os.path.join(tmp_dir_to_dump, "Simple_CNN1.onnx")) if (caslib is not None) and tmp_caslib: self.s.retrieve('table.dropcaslib', message_level='error', caslib=caslib)
def initial_block(inp): ''' Defines the initial block of ENet Parameters ---------- inp : class:`InputLayer` Input layer Returns ------- :class:`Concat` ''' x = Conv2d(13, 3, stride=2, padding=1, act='identity', include_bias=False)(inp) x_bn = BN(act='relu')(x) y = Pooling(2)(inp) merge = Concat()([x_bn, y]) return merge
def YoloV2_MultiSize(conn, anchors, model_table='YoloV2-MultiSize', n_channels=3, width=416, height=416, scale=1.0 / 255, random_mutation=None, act='leaky', act_detection='AUTO', softmax_for_class_prob=True, coord_type='YOLO', max_label_per_image=30, max_boxes=30, n_classes=20, predictions_per_grid=5, do_sqrt=True, grid_number=13, coord_scale=None, object_scale=None, prediction_not_a_object_scale=None, class_scale=None, detection_threshold=None, iou_threshold=None, random_boxes=False, match_anchor_size=None, num_to_force_coord=None, random_flip=None, random_crop=None): ''' Generates a deep learning model with the Yolov2 architecture. The model is same as Yolov2 proposed in original paper. In addition to Yolov2, the model adds a passthrough layer that brings feature from an earlier layer to lower resolution layer. Parameters ---------- conn : CAS Specifies the connection of the CAS connection. anchors : list Specifies the anchor box values. model_table : string, optional Specifies the name of CAS table to store the model. n_channels : int, optional Specifies the number of the channels (i.e., depth) of the input layer. Default: 3 width : int, optional Specifies the width of the input layer. Default: 416 height : int, optional Specifies the height of the input layer. Default: 416 scale : double, optional Specifies a scaling factor to be applied to each pixel intensity values. Default: 1.0 / 255 random_mutation : string, optional Specifies how to apply data augmentations/mutations to the data in the input layer. Valid Values: 'none', 'random' act : string, optional Specifies the activation function for the batch normalization layers. Default: 'leaky' act_detection : string, optional Specifies the activation function for the detection layer. Valid Values: AUTO, IDENTITY, LOGISTIC, SIGMOID, TANH, RECTIFIER, RELU, SOFPLUS, ELU, LEAKY, FCMP Default: AUTO softmax_for_class_prob : bool, optional Specifies whether to perform Softmax on class probability per predicted object. Default: True coord_type : string, optional Specifies the format of how to represent bounding boxes. For example, a bounding box can be represented with the x and y locations of the top-left point as well as width and height of the rectangle. This format is the 'rect' format. We also support coco and yolo formats. Valid Values: 'rect', 'yolo', 'coco' Default: 'yolo' max_label_per_image : int, optional Specifies the maximum number of labels per image in the training. Default: 30 max_boxes : int, optional Specifies the maximum number of overall predictions allowed in the detection layer. Default: 30 n_classes : int, optional Specifies the number of classes. If None is assigned, the model will automatically detect the number of classes based on the training set. Default: 20 predictions_per_grid : int, optional Specifies the amount of predictions will be done per grid. Default: 5 do_sqrt : bool, optional Specifies whether to apply the SQRT function to width and height of the object for the cost function. Default: True grid_number : int, optional Specifies the amount of cells to be analyzed for an image. For example, if the value is 5, then the image will be divided into a 5 x 5 grid. Default: 13 coord_scale : float, optional Specifies the weight for the cost function in the detection layer, when objects exist in the grid. object_scale : float, optional Specifies the weight for object detected for the cost function in the detection layer. prediction_not_a_object_scale : float, optional Specifies the weight for the cost function in the detection layer, when objects do not exist in the grid. class_scale : float, optional Specifies the weight for the class of object detected for the cost function in the detection layer. detection_threshold : float, optional Specifies the threshold for object detection. iou_threshold : float, optional Specifies the IOU Threshold of maximum suppression in object detection. random_boxes : bool, optional Randomizing boxes when loading the bounding box information. Default: False match_anchor_size : bool, optional Whether to force the predicted box match the anchor boxes in sizes for all predictions num_to_force_coord : int, optional The number of leading chunk of images in training when the algorithm forces predicted objects in each grid to be equal to the anchor box sizes, and located at the grid center random_flip : string, optional Specifies how to flip the data in the input layer when image data is used. Approximately half of the input data is subject to flipping. Valid Values: 'h', 'hv', 'v', 'none' random_crop : string, optional Specifies how to crop the data in the input layer when image data is used. Images are cropped to the values that are specified in the width and height parameters. Only the images with one or both dimensions that are larger than those sizes are cropped. Valid Values: 'none', 'unique', 'randomresized', 'resizethencrop' Returns ------- :class:`Sequential` References ---------- https://arxiv.org/pdf/1612.08242.pdf ''' model = Sequential(conn=conn, model_table=model_table) parameters = locals() input_parameters = get_layer_options(input_layer_options, parameters) model.add(InputLayer(**input_parameters)) # conv1 224 416 model.add(Conv2d(32, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) model.add(Pooling(width=2, height=2, stride=2, pool='max')) # conv2 112 208 model.add(Conv2d(64, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) model.add(Pooling(width=2, height=2, stride=2, pool='max')) # conv3 56 104 model.add( Conv2d(128, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv4 56 104 model.add(Conv2d(64, width=1, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv5 56 104 model.add( Conv2d(128, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) model.add(Pooling(width=2, height=2, stride=2, pool='max')) # conv6 28 52 model.add( Conv2d(256, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv7 28 52 model.add( Conv2d(128, width=1, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv8 28 52 model.add( Conv2d(256, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) model.add(Pooling(width=2, height=2, stride=2, pool='max')) # conv9 14 26 model.add( Conv2d(512, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv10 14 26 model.add( Conv2d(256, width=1, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv11 14 26 model.add( Conv2d(512, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv12 14 26 model.add( Conv2d(256, width=1, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv13 14 26 model.add( Conv2d(512, width=3, act='identity', include_bias=False, stride=1)) pointLayer1 = BN(act=act, name='BN5_13') model.add(pointLayer1) model.add(Pooling(width=2, height=2, stride=2, pool='max')) # conv14 7 13 model.add( Conv2d(1024, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv15 7 13 model.add( Conv2d(512, width=1, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv16 7 13 model.add( Conv2d(1024, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv17 7 13 model.add( Conv2d(512, width=1, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv18 7 13 model.add( Conv2d(1024, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) # conv19 7 13 model.add( Conv2d(1024, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act, name='BN6_19')) # conv20 7 13 model.add( Conv2d(1024, width=3, act='identity', include_bias=False, stride=1)) pointLayer2 = BN(act=act, name='BN6_20') model.add(pointLayer2) # conv21 7 26 * 26 * 512 -> 26 * 26 * 64 model.add( Conv2d(64, width=1, act='identity', include_bias=False, stride=1, src_layers=[pointLayer1])) model.add(BN(act=act)) # reshape 26 * 26 * 64 -> 13 * 13 * 256 pointLayer3 = Reshape(act='identity', width=grid_number, height=grid_number, depth=256, name='reshape1') model.add(pointLayer3) # concat model.add(Concat(act='identity', src_layers=[pointLayer2, pointLayer3])) # conv22 7 13 model.add( Conv2d(1024, width=3, act='identity', include_bias=False, stride=1)) model.add(BN(act=act)) model.add( Conv2d((n_classes + 5) * predictions_per_grid, width=1, act='identity', include_bias=False, stride=1)) model.add( Detection(act=act_detection, detection_model_type='yolov2', anchors=anchors, softmax_for_class_prob=softmax_for_class_prob, coord_type=coord_type, class_number=n_classes, grid_number=grid_number, predictions_per_grid=predictions_per_grid, do_sqrt=do_sqrt, coord_scale=coord_scale, object_scale=object_scale, prediction_not_a_object_scale=prediction_not_a_object_scale, class_scale=class_scale, detection_threshold=detection_threshold, iou_threshold=iou_threshold, random_boxes=random_boxes, max_label_per_image=max_label_per_image, max_boxes=max_boxes, match_anchor_size=match_anchor_size, num_to_force_coord=num_to_force_coord)) return model
def DenseNet121(conn, model_table='DENSENET121', n_classes=1000, conv_channel=64, growth_rate=32, n_cells=[6, 12, 24, 16], n_channels=3, reduction=0.5, width=224, height=224, scale=1, random_flip=None, random_crop=None, offsets=(103.939, 116.779, 123.68), random_mutation=None): ''' Generates a deep learning model with the DenseNet121 architecture. Parameters ---------- conn : CAS Specifies the connection of the CAS connection. model_table : string Specifies the name of CAS table to store the model. n_classes : int, optional Specifies the number of classes. If None is assigned, the model will automatically detect the number of classes based on the training set. Default: 1000 conv_channel : int, optional Specifies the number of filters of the first convolution layer. Default: 64 growth_rate : int, optional Specifies the growth rate of convolution layers. Default: 32 n_cells : int array length=4, optional Specifies the number of dense connection for each DenseNet block. Default: [6, 12, 24, 16] reduction : double, optional Specifies the factor of transition blocks. Default: 0.5 n_channels : int, optional Specifies the number of the channels (i.e., depth) of the input layer. Default: 3. width : int, optional Specifies the width of the input layer. Default: 224. height : int, optional Specifies the height of the input layer. Default: 224. scale : double, optional Specifies a scaling factor to be applied to each pixel intensity values. Default: 1. random_flip : string, optional Specifies how to flip the data in the input layer when image data is used. Approximately half of the input data is subject to flipping. Valid Values: 'h', 'hv', 'v', 'none' random_crop : string, optional Specifies how to crop the data in the input layer when image data is used. Images are cropped to the values that are specified in the width and height parameters. Only the images with one or both dimensions that are larger than those sizes are cropped. Valid Values: 'none', 'unique', 'randomresized', 'resizethencrop' offsets : double or iter-of-doubles, optional Specifies an offset for each channel in the input data. The final input data is set after applying scaling and subtracting the specified offsets. Default: (103.939, 116.779, 123.68) random_mutation : string, optional Specifies how to apply data augmentations/mutations to the data in the input layer. Valid Values: 'none', 'random' Returns ------- :class:`Sequential` References ---------- https://arxiv.org/pdf/1608.06993.pdf ''' conn.retrieve('loadactionset', _messagelevel='error', actionset='deeplearn') # get all the parms passed in parameters = locals() n_blocks = len(n_cells) model = Sequential(conn=conn, model_table=model_table) # get the input parameters input_parameters = get_layer_options(input_layer_options, parameters) model.add(InputLayer(**input_parameters)) # Top layers model.add( Conv2d(conv_channel, width=7, act='identity', include_bias=False, stride=2)) model.add(BN(act='relu')) src_layer = Pooling(width=3, height=3, stride=2, padding=1, pool='max') model.add(src_layer) for i in range(n_blocks): for _ in range(n_cells[i]): model.add(BN(act='relu')) model.add( Conv2d(n_filters=growth_rate * 4, width=1, act='identity', stride=1, include_bias=False)) model.add(BN(act='relu')) src_layer2 = Conv2d(n_filters=growth_rate, width=3, act='identity', stride=1, include_bias=False) model.add(src_layer2) src_layer = Concat(act='identity', src_layers=[src_layer, src_layer2]) model.add(src_layer) conv_channel += growth_rate if i != (n_blocks - 1): # transition block conv_channel = int(conv_channel * reduction) model.add(BN(act='relu')) model.add( Conv2d(n_filters=conv_channel, width=1, act='identity', stride=1, include_bias=False)) src_layer = Pooling(width=2, height=2, stride=2, pool='mean') model.add(src_layer) model.add(BN(act='identity')) # Bottom Layers model.add(GlobalAveragePooling2D()) model.add(OutputLayer(act='softmax', n=n_classes)) return model
def _shuffle_unit(inputs, in_channels, out_channels, groups, bottleneck_ratio, strides=2, stage=1, block=1): """ create a shuffle unit Parameters ---------- inputs: Input tensor of with `channels_last` data format in_channels: number of input channels out_channels: number of output channels strides: An integer or tuple/list of 2 integers, groups: number of groups per channel bottleneck_ratio: float bottleneck ratio implies the ratio of bottleneck channels to output channels. stage: stage number block: block number """ prefix = 'stage%d/block%d' % (stage, block) # if strides >= 2: # out_channels -= in_channels # default: 1/4 of the output channel of a ShuffleNet Unit bottleneck_channels = int(out_channels * bottleneck_ratio) groups = (1 if stage == 2 and block == 1 else groups) # x = _group_conv(inputs, in_channels, out_channels = bottleneck_channels, # groups = (1 if stage == 2 and block == 1 else groups), # name = '%s/1x1_gconv_1' % prefix) x = GroupConv2d(bottleneck_channels, n_groups=(1 if stage == 2 and block == 1 else groups), act='identity', width=1, height=1, stride=1, include_bias=False, name='%s/1x1_gconv_1' % prefix)(inputs) x = BN(act='relu', name='%s/bn_gconv_1' % prefix)(x) x = ChannelShuffle(n_groups=groups, name='%s/channel_shuffle' % prefix)(x) # depthwise convolutioin x = GroupConv2d(x.shape[-1], n_groups=x.shape[-1], width=3, height=3, include_bias=False, stride=strides, act='identity', name='%s/1x1_dwconv_1' % prefix)(x) x = BN(act=block_act, name='%s/bn_dwconv_1' % prefix)(x) out_channels = out_channels if strides == 1 else out_channels - in_channels x = GroupConv2d(out_channels, n_groups=groups, width=1, height=1, stride=1, act='identity', include_bias=False, name='%s/1x1_gconv_2' % prefix)(x) x = BN(act=block_act, name='%s/bn_gconv_2' % prefix)(x) if strides < 2: ret = Res(act='relu', name='%s/add' % prefix)([x, inputs]) else: avg = Pooling(width=3, height=3, stride=2, pool='mean', name='%s/avg_pool' % prefix)(inputs) ret = Concat(act='relu', name='%s/concat' % prefix)([x, avg]) return ret
def UNet(conn, model_table='UNet', n_classes=2, n_channels=1, width=256, height=256, scale=1.0 / 255, norm_stds=None, offsets=None, random_mutation=None, init=None, bn_after_convolutions=False, random_flip=None, random_crop=None): ''' Generates a deep learning model with the U-Net architecture. Parameters ---------- conn : CAS Specifies the connection of the CAS connection. model_table : string, optional Specifies the name of CAS table to store the model. n_classes : int, optional Specifies the number of classes. If None is assigned, the model will automatically detect the number of classes based on the training set. Default: 2 n_channels : int, optional Specifies the number of the channels (i.e., depth) of the input layer. Default: 3 width : int, optional Specifies the width of the input layer. Default: 256 height : int, optional Specifies the height of the input layer. Default: 256 scale : double, optional Specifies a scaling factor to be applied to each pixel intensity values. Default: 1.0/255 norm_stds : double or iter-of-doubles, optional Specifies a standard deviation for each channel in the input data. The final input data is normalized with specified means and standard deviations. offsets : double or iter-of-doubles, optional Specifies an offset for each channel in the input data. The final input data is set after applying scaling and subtracting the specified offsets. random_mutation : string, optional Specifies how to apply data augmentations/mutations to the data in the input layer. Valid Values: 'none', 'random' init : str Specifies the initialization scheme for convolution layers. Valid Values: XAVIER, UNIFORM, NORMAL, CAUCHY, XAVIER1, XAVIER2, MSRA, MSRA1, MSRA2 Default: None bn_after_convolutions : Boolean If set to True, a batch normalization layer is added after each convolution layer. random_flip : string, optional Specifies how to flip the data in the input layer when image data is used. Approximately half of the input data is subject to flipping. Valid Values: 'h', 'hv', 'v', 'none' random_crop : string, optional Specifies how to crop the data in the input layer when image data is used. Images are cropped to the values that are specified in the width and height parameters. Only the images with one or both dimensions that are larger than those sizes are cropped. Valid Values: 'none', 'unique', 'randomresized', 'resizethencrop' Returns ------- :class:`Sequential` References ---------- https://arxiv.org/pdf/1505.04597 ''' parameters = locals() input_parameters = get_layer_options(input_layer_options, parameters) inp = Input(**input_parameters, name='data') act_conv = 'relu' bias_conv = True if bn_after_convolutions: act_conv = 'identity' bias_conv = False # The model follows UNet paper architecture. The network down-samples by performing max pooling with stride=2 conv1 = Conv2d(64, 3, act=act_conv, init=init, include_bias=bias_conv)(inp) conv1 = BN(act='relu')(conv1) if bn_after_convolutions else conv1 conv1 = Conv2d(64, 3, act=act_conv, init=init, include_bias=bias_conv)(conv1) conv1 = BN(act='relu')(conv1) if bn_after_convolutions else conv1 pool1 = Pooling(2)(conv1) conv2 = Conv2d(128, 3, act=act_conv, init=init, include_bias=bias_conv)(pool1) conv2 = BN(act='relu')(conv2) if bn_after_convolutions else conv2 conv2 = Conv2d(128, 3, act=act_conv, init=init, include_bias=bias_conv)(conv2) conv2 = BN(act='relu')(conv2) if bn_after_convolutions else conv2 pool2 = Pooling(2)(conv2) conv3 = Conv2d(256, 3, act=act_conv, init=init, include_bias=bias_conv)(pool2) conv3 = BN(act='relu')(conv3) if bn_after_convolutions else conv3 conv3 = Conv2d(256, 3, act=act_conv, init=init, include_bias=bias_conv)(conv3) conv3 = BN(act='relu')(conv3) if bn_after_convolutions else conv3 pool3 = Pooling(2)(conv3) conv4 = Conv2d(512, 3, act=act_conv, init=init, include_bias=bias_conv)(pool3) conv4 = BN(act='relu')(conv4) if bn_after_convolutions else conv4 conv4 = Conv2d(512, 3, act=act_conv, init=init, include_bias=bias_conv)(conv4) conv4 = BN(act='relu')(conv4) if bn_after_convolutions else conv4 pool4 = Pooling(2)(conv4) conv5 = Conv2d(1024, 3, act=act_conv, init=init, include_bias=bias_conv)(pool4) conv5 = BN(act='relu')(conv5) if bn_after_convolutions else conv5 conv5 = Conv2d(1024, 3, act=act_conv, init=init, include_bias=bias_conv)(conv5) conv5 = BN(act='relu')(conv5) if bn_after_convolutions else conv5 # the minimum is 1/2^4 of the original image size # Our implementation applies Transpose convolution to upsample feature maps. tconv6 = Conv2DTranspose(512, 3, stride=2, act='relu', padding=1, output_size=conv4.shape, init=init)(conv5) # 64 # concatenation layers to combine encoder and decoder features merge6 = Concat()([conv4, tconv6]) conv6 = Conv2d(512, 3, act=act_conv, init=init, include_bias=bias_conv)(merge6) conv6 = BN(act='relu')(conv6) if bn_after_convolutions else conv6 conv6 = Conv2d(512, 3, act=act_conv, init=init, include_bias=bias_conv)(conv6) conv6 = BN(act='relu')(conv6) if bn_after_convolutions else conv6 tconv7 = Conv2DTranspose(256, 3, stride=2, act='relu', padding=1, output_size=conv3.shape, init=init)(conv6) # 128 merge7 = Concat()([conv3, tconv7]) conv7 = Conv2d(256, 3, act=act_conv, init=init, include_bias=bias_conv)(merge7) conv7 = BN(act='relu')(conv7) if bn_after_convolutions else conv7 conv7 = Conv2d(256, 3, act=act_conv, init=init, include_bias=bias_conv)(conv7) conv7 = BN(act='relu')(conv7) if bn_after_convolutions else conv7 tconv8 = Conv2DTranspose(128, stride=2, act='relu', padding=1, output_size=conv2.shape, init=init)(conv7) # 256 merge8 = Concat()([conv2, tconv8]) conv8 = Conv2d(128, 3, act=act_conv, init=init, include_bias=bias_conv)(merge8) conv8 = BN(act='relu')(conv8) if bn_after_convolutions else conv8 conv8 = Conv2d(128, 3, act=act_conv, init=init, include_bias=bias_conv)(conv8) conv8 = BN(act='relu')(conv8) if bn_after_convolutions else conv8 tconv9 = Conv2DTranspose(64, stride=2, act='relu', padding=1, output_size=conv1.shape, init=init)(conv8) # 512 merge9 = Concat()([conv1, tconv9]) conv9 = Conv2d(64, 3, act=act_conv, init=init, include_bias=bias_conv)(merge9) conv9 = BN(act='relu')(conv9) if bn_after_convolutions else conv9 conv9 = Conv2d(64, 3, act=act_conv, init=init, include_bias=bias_conv)(conv9) conv9 = BN(act='relu')(conv9) if bn_after_convolutions else conv9 conv9 = Conv2d(n_classes, 3, act='relu', init=init)(conv9) seg1 = Segmentation(name='Segmentation_1')(conv9) model = Model(conn, inputs=inp, outputs=seg1, model_table=model_table) model.compile() return model
def InceptionV3(conn, model_table='InceptionV3', n_classes=1000, n_channels=3, width=299, height=299, scale=1, random_flip=None, random_crop=None, offsets=(103.939, 116.779, 123.68), pre_trained_weights=False, pre_trained_weights_file=None, include_top=False, random_mutation=None): ''' Generates a deep learning model with the Inceptionv3 architecture with batch normalization layers. Parameters ---------- conn : CAS Specifies the CAS connection object. model_table : string, optional Specifies the name of CAS table to store the model in. n_classes : int, optional Specifies the number of classes. If None is assigned, the model will automatically detect the number of classes based on the training set. Default: 1000 n_channels : int, optional Specifies the number of the channels (i.e., depth) of the input layer. Default: 3 width : int, optional Specifies the width of the input layer. Default: 299 height : int, optional Specifies the height of the input layer. Default: 299 scale : double, optional Specifies a scaling factor to be applied to each pixel intensity values. Default: 1.0 random_flip : string, optional Specifies how to flip the data in the input layer when image data is used. Approximately half of the input data is subject to flipping. Valid Values: 'h', 'hv', 'v', 'none' random_crop : string, optional Specifies how to crop the data in the input layer when image data is used. Images are cropped to the values that are specified in the width and height parameters. Only the images with one or both dimensions that are larger than those sizes are cropped. Valid Values: 'none', 'unique', 'randomresized', 'resizethencrop' offsets : double or iter-of-doubles, optional Specifies an offset for each channel in the input data. The final input data is set after applying scaling and subtracting the specified offsets. Default: (103.939, 116.779, 123.68) pre_trained_weights : bool, optional Specifies whether to use the pre-trained weights from ImageNet data set Default: False pre_trained_weights_file : string, optional Specifies the file name for the pretained weights. Must be a fully qualified file name of SAS-compatible file (*.caffemodel.h5) Note: Required when pre_train_weight=True. include_top : bool, optional Specifies whether to include pre-trained weights of the top layers, i.e. the FC layers Default: False random_mutation : string, optional Specifies how to apply data augmentations/mutations to the data in the input layer. Valid Values: 'none', 'random' Returns ------- :class:`Sequential` If `pre_train_weight` is `False` :class:`Model` If `pre_train_weight` is `True` References ---------- https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Szegedy_Rethinking_the_Inception_CVPR_2016_paper.pdf ''' conn.retrieve('loadactionset', _messagelevel='error', actionset='deeplearn') # get all the parms passed in parameters = locals() if not pre_trained_weights: model = Sequential(conn=conn, model_table=model_table) # get the input parameters input_parameters = get_layer_options(input_layer_options, parameters) model.add(InputLayer(**input_parameters)) # 299 x 299 x 3 model.add( Conv2d(n_filters=32, width=3, height=3, stride=2, act='identity', include_bias=False, padding=0)) model.add(BN(act='relu')) # 149 x 149 x 32 model.add( Conv2d(n_filters=32, width=3, height=3, stride=1, act='identity', include_bias=False, padding=0)) model.add(BN(act='relu')) # 147 x 147 x 32 model.add( Conv2d(n_filters=64, width=3, height=3, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) # 147 x 147 x 64 model.add(Pooling(width=3, height=3, stride=2, pool='max', padding=0)) # 73 x 73 x 64 model.add( Conv2d(n_filters=80, width=1, height=1, stride=1, act='identity', include_bias=False, padding=0)) model.add(BN(act='relu')) # 73 x 73 x 80 model.add( Conv2d(n_filters=192, width=3, height=3, stride=1, act='identity', include_bias=False, padding=0)) model.add(BN(act='relu')) # 71 x 71 x 192 pool2 = Pooling(width=3, height=3, stride=2, pool='max', padding=0) model.add(pool2) # mixed 0: output 35 x 35 x 256 # branch1x1 model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[pool2])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch5x5 model.add( Conv2d(n_filters=48, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[pool2])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=64, width=5, height=5, stride=1, act='identity', include_bias=False)) branch5x5 = BN(act='relu') model.add(branch5x5) # branch3x3dbl model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[pool2])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) branch3x3dbl = BN(act='relu') model.add(branch3x3dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[pool2])) model.add( Conv2d(n_filters=32, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # mixed0 concat concat = Concat( act='identity', src_layers=[branch1x1, branch5x5, branch3x3dbl, branch_pool]) model.add(concat) # mixed 1: output 35 x 35 x 288 # branch1x1 model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch5x5 model.add( Conv2d(n_filters=48, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=64, width=5, height=5, stride=1, act='identity', include_bias=False)) branch5x5 = BN(act='relu') model.add(branch5x5) # branch3x3dbl model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) branch3x3dbl = BN(act='relu') model.add(branch3x3dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[concat])) model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # mixed1 concat concat = Concat( act='identity', src_layers=[branch1x1, branch5x5, branch3x3dbl, branch_pool]) model.add(concat) # mixed 2: output 35 x 35 x 288 # branch1x1 model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch5x5 model.add( Conv2d(n_filters=48, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=64, width=5, height=5, stride=1, act='identity', include_bias=False)) branch5x5 = BN(act='relu') model.add(branch5x5) # branch3x3dbl model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) branch3x3dbl = BN(act='relu') model.add(branch3x3dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[concat])) model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # mixed2 concat concat = Concat( act='identity', src_layers=[branch1x1, branch5x5, branch3x3dbl, branch_pool]) model.add(concat) # mixed 3: output 17 x 17 x 768 # branch3x3 model.add( Conv2d(n_filters=384, width=3, height=3, stride=2, act='identity', include_bias=False, padding=0, src_layers=[concat])) branch3x3 = BN(act='relu') model.add(branch3x3) # branch3x3dbl model.add( Conv2d(n_filters=64, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=96, width=3, height=3, stride=2, act='identity', include_bias=False, padding=0)) branch3x3dbl = BN(act='relu') model.add(branch3x3dbl) # branch_pool branch_pool = Pooling(width=3, height=3, stride=2, pool='max', padding=0, src_layers=[concat]) model.add(branch_pool) # mixed3 concat concat = Concat(act='identity', src_layers=[branch3x3, branch3x3dbl, branch_pool]) model.add(concat) # mixed 4: output 17 x 17 x 768 # branch1x1 model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch7x7 model.add( Conv2d(n_filters=128, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=128, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=1, height=7, stride=1, act='identity', include_bias=False)) branch7x7 = BN(act='relu') model.add(branch7x7) # branch7x7dbl model.add( Conv2d(n_filters=128, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=128, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=128, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=128, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=7, height=1, stride=1, act='identity', include_bias=False)) branch7x7dbl = BN(act='relu') model.add(branch7x7dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[concat])) model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # mixed4 concat concat = Concat( act='identity', src_layers=[branch1x1, branch7x7, branch7x7dbl, branch_pool]) model.add(concat) # mixed 5, 6: output 17 x 17 x 768 for i in range(2): # branch1x1 model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch7x7 model.add( Conv2d(n_filters=160, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=160, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=1, height=7, stride=1, act='identity', include_bias=False)) branch7x7 = BN(act='relu') model.add(branch7x7) # branch7x7dbl model.add( Conv2d(n_filters=160, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=160, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=160, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=160, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=7, height=1, stride=1, act='identity', include_bias=False)) branch7x7dbl = BN(act='relu') model.add(branch7x7dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[concat])) model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # concat concat = Concat( act='identity', src_layers=[branch1x1, branch7x7, branch7x7dbl, branch_pool]) model.add(concat) # mixed 7: output 17 x 17 x 768 # branch1x1 model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch7x7 model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=1, height=7, stride=1, act='identity', include_bias=False)) branch7x7 = BN(act='relu') model.add(branch7x7) # branch7x7dbl model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=7, height=1, stride=1, act='identity', include_bias=False)) branch7x7dbl = BN(act='relu') model.add(branch7x7dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[concat])) model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # mixed7 concat concat = Concat( act='identity', src_layers=[branch1x1, branch7x7, branch7x7dbl, branch_pool]) model.add(concat) # mixed 8: output 8 x 8 x 1280 # branch3x3 model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=320, width=3, height=3, stride=2, act='identity', include_bias=False, padding=0)) branch3x3 = BN(act='relu') model.add(branch3x3) # branch7x7x3 model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=7, height=1, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=1, height=7, stride=1, act='identity', include_bias=False)) model.add(BN(act='relu')) model.add( Conv2d(n_filters=192, width=3, height=3, stride=2, act='identity', include_bias=False, padding=0)) branch7x7x3 = BN(act='relu') model.add(branch7x7x3) # branch_pool branch_pool = Pooling(width=3, height=3, stride=2, pool='max', padding=0, src_layers=[concat]) model.add(branch_pool) # mixed8 concat concat = Concat(act='identity', src_layers=[branch3x3, branch7x7x3, branch_pool]) model.add(concat) # mixed 9, 10: output 8 x 8 x 2048 for i in range(2): # branch1x1 model.add( Conv2d(n_filters=320, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch1x1 = BN(act='relu') model.add(branch1x1) # branch3x3 model.add( Conv2d(n_filters=384, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) branch3x3 = BN(act='relu') model.add(branch3x3) model.add( Conv2d(n_filters=384, width=3, height=1, stride=1, act='identity', include_bias=False, src_layers=[branch3x3])) branch3x3_1 = BN(act='relu') model.add(branch3x3_1) model.add( Conv2d(n_filters=384, width=1, height=3, stride=1, act='identity', include_bias=False, src_layers=[branch3x3])) branch3x3_2 = BN(act='relu') model.add(branch3x3_2) branch3x3 = Concat(act='identity', src_layers=[branch3x3_1, branch3x3_2]) model.add(branch3x3) # branch3x3dbl model.add( Conv2d(n_filters=448, width=1, height=1, stride=1, act='identity', include_bias=False, src_layers=[concat])) model.add(BN(act='relu')) model.add( Conv2d(n_filters=384, width=3, height=3, stride=1, act='identity', include_bias=False)) branch3x3dbl = BN(act='relu') model.add(branch3x3dbl) model.add( Conv2d(n_filters=384, width=3, height=1, stride=1, act='identity', include_bias=False, src_layers=[branch3x3dbl])) branch3x3dbl_1 = BN(act='relu') model.add(branch3x3dbl_1) model.add( Conv2d(n_filters=384, width=1, height=3, stride=1, act='identity', include_bias=False, src_layers=[branch3x3dbl])) branch3x3dbl_2 = BN(act='relu') model.add(branch3x3dbl_2) branch3x3dbl = Concat(act='identity', src_layers=[branch3x3dbl_1, branch3x3dbl_2]) model.add(branch3x3dbl) # branch_pool model.add( Pooling(width=3, height=3, stride=1, pool='average', src_layers=[concat])) model.add( Conv2d(n_filters=192, width=1, height=1, stride=1, act='identity', include_bias=False)) branch_pool = BN(act='relu') model.add(branch_pool) # concat concat = Concat( act='identity', src_layers=[branch1x1, branch3x3, branch3x3dbl, branch_pool]) model.add(concat) # calculate dimensions for global average pooling w = max((width - 75) // 32 + 1, 1) h = max((height - 75) // 32 + 1, 1) # global average pooling model.add( Pooling(width=w, height=h, stride=1, pool='average', padding=0, src_layers=[concat])) # output layer model.add(OutputLayer(n=n_classes)) return model else: if pre_trained_weights_file is None: raise ValueError( '\nThe pre-trained weights file is not specified.\n' 'Please follow the steps below to attach the ' 'pre-trained weights:\n' '1. Go to the website ' 'https://support.sas.com/documentation/prod-p/vdmml/zip/ ' 'and download the associated weight file.\n' '2. Upload the *.h5 file to ' 'a server side directory which the CAS ' 'session has access to.\n' '3. Specify the pre_train_weight_file using ' 'the fully qualified server side path.') print('NOTE: Scale is set to 1/127.5, and offsets 1 to ' 'match Keras preprocessing.') model_cas = model_inceptionv3.InceptionV3_Model( s=conn, model_table=model_table, n_channels=n_channels, width=width, height=height, random_crop=random_crop, offsets=[1, 1, 1], random_flip=random_flip, random_mutation=random_mutation) if include_top: if n_classes != 1000: warnings.warn( 'If include_top = True, ' 'n_classes will be set to 1000.', RuntimeWarning) model = Model.from_table(model_cas) model.load_weights(path=pre_trained_weights_file, labels=True) return model else: model = Model.from_table(model_cas, display_note=False) model.load_weights(path=pre_trained_weights_file) weight_table_options = model.model_weights.to_table_params() weight_table_options.update(dict(where='_LayerID_<218')) model._retrieve_('table.partition', table=weight_table_options, casout=dict( replace=True, **model.model_weights.to_table_params())) model._retrieve_('deeplearn.removelayer', model=model_table, name='predictions') model._retrieve_('deeplearn.addlayer', model=model_table, name='predictions', layer=dict(type='output', n=n_classes, act='softmax'), srcLayers=['avg_pool']) model = Model.from_table(conn.CASTable(model_table)) return model