Ssd_keras: Error when checking model target

[fortunto2]

Hi, nice work....
but i have errror in notebook ssd
please help ;)

Exception: Error when checking : expected input_2 to have shape (None, 300, 300, 3) but got array with shape (5, 3, 300, 300)

python 2.7 anaconda
keras and theano,tf updated to last

[fortunto2]

in python 3 warning

/home/rustam/anaconda2/envs/py3k/lib/python3.5/site-packages/keras/backend/theano_backend.py:1314: UserWarning: dict_keys(['filter_dilation']) are now deprecated in `tensor.nnet.abstract_conv.conv2d` interface and will be ignored.
  filter_dilation=filter_dilation)

[MrXu]

@fortunto2 I think the example is using tensorflow. In tensorflow, image converted to array has a different sequence from the one in theano. You can try to use tensorflow backend.

[rykov8]

Alternatively, you can try to pass (3, 300, 300) as input_shape to SSD while constructing the net. However, Theano backend was not tested (though I tried to support it for forward pass) and, moreover, the weights are also in Tensorflow format, so, using Theano without weights' conversion will be quite slow. What is more important, the loss function is implemented only in Tensorflow, so, I agree with @MrXu and also suggest you to switch backend.

[fortunto2]

ok thanks, not principal...
but in tensorflow i catch new error))

ValueError                                Traceback (most recent call last)
<ipython-input-3-ea636890c378> in <module>()
      1 input_shape=(300, 300, 3)
----> 2 model = SSD(input_shape, num_classes=NUM_CLASSES)
      3 model.load_weights('weights_300x300.hdf5', by_name=True)
      4 bbox_util = BBoxUtility(NUM_CLASSES)

/opt/neural/ssd_keras/ssd.pyc in SSD(input_shape, num_classes)
     40                                    activation='relu',
     41                                    border_mode='same',
---> 42                                    name='conv2_1')(net['pool1'])
     43     net['conv2_2'] = Convolution2D(128, 3, 3,
     44                                    activation='relu',

/home/rustam/anaconda2/lib/python2.7/site-packages/keras/engine/topology.pyc in __call__(self, x, mask)
    514         if inbound_layers:
    515             # This will call layer.build() if necessary.
--> 516             self.add_inbound_node(inbound_layers, node_indices, tensor_indices)
    517             # Outputs were already computed when calling self.add_inbound_node.
    518             outputs = self.inbound_nodes[-1].output_tensors

/home/rustam/anaconda2/lib/python2.7/site-packages/keras/engine/topology.pyc in add_inbound_node(self, inbound_layers, node_indices, tensor_indices)
    568         # creating the node automatically updates self.inbound_nodes
    569         # as well as outbound_nodes on inbound layers.
--> 570         Node.create_node(self, inbound_layers, node_indices, tensor_indices)
    571 
    572     def get_output_shape_for(self, input_shape):

/home/rustam/anaconda2/lib/python2.7/site-packages/keras/engine/topology.pyc in create_node(cls, outbound_layer, inbound_layers, node_indices, tensor_indices)
    152 
    153         if len(input_tensors) == 1:
--> 154             output_tensors = to_list(outbound_layer.call(input_tensors[0], mask=input_masks[0]))
    155             output_masks = to_list(outbound_layer.compute_mask(input_tensors[0], input_masks[0]))
    156             # TODO: try to auto-infer shape if exception is raised by get_output_shape_for.

/home/rustam/anaconda2/lib/python2.7/site-packages/keras/layers/convolutional.pyc in call(self, x, mask)
    464                           border_mode=self.border_mode,
    465                           dim_ordering=self.dim_ordering,
--> 466                           filter_shape=self.W_shape)
    467         if self.bias:
    468             if self.dim_ordering == 'th':

/home/rustam/anaconda2/lib/python2.7/site-packages/keras/backend/tensorflow_backend.pyc in conv2d(x, kernel, strides, border_mode, dim_ordering, image_shape, filter_shape, filter_dilation)
   1675     if filter_dilation == (1, 1):
   1676         strides = (1,) + strides + (1,)
-> 1677         x = tf.nn.conv2d(x, kernel, strides, padding=padding)
   1678     else:
   1679         assert filter_dilation[0] == filter_dilation[1]

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/gen_nn_ops.pyc in conv2d(input, filter, strides, padding, use_cudnn_on_gpu, data_format, name)
    392                                 strides=strides, padding=padding,
    393                                 use_cudnn_on_gpu=use_cudnn_on_gpu,
--> 394                                 data_format=data_format, name=name)
    395   return result
    396 

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.pyc in apply_op(self, op_type_name, name, **keywords)
    701           op = g.create_op(op_type_name, inputs, output_types, name=scope,
    702                            input_types=input_types, attrs=attr_protos,
--> 703                            op_def=op_def)
    704           outputs = op.outputs
    705           return _Restructure(ops.convert_n_to_tensor(outputs),

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.pyc in create_op(self, op_type, inputs, dtypes, input_types, name, attrs, op_def, compute_shapes, compute_device)
   2310                     original_op=self._default_original_op, op_def=op_def)
   2311     if compute_shapes:
-> 2312       set_shapes_for_outputs(ret)
   2313     self._add_op(ret)
   2314     self._record_op_seen_by_control_dependencies(ret)

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/ops.pyc in set_shapes_for_outputs(op)
   1702       raise RuntimeError("No shape function registered for standard op: %s"
   1703                          % op.type)
-> 1704   shapes = shape_func(op)
   1705   if shapes is None:
   1706     raise RuntimeError(

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.pyc in conv2d_shape(op)
    244   out_rows, out_cols = get2d_conv_output_size(in_rows, in_cols, filter_rows,
    245                                               filter_cols, stride_r, stride_c,
--> 246                                               padding)
    247 
    248   output_shape = [batch_size, out_rows, out_cols, depth_out]

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.pyc in get2d_conv_output_size(input_height, input_width, filter_height, filter_width, row_stride, col_stride, padding_type)
    182   return get_conv_output_size((input_height, input_width),
    183                               (filter_height, filter_width),
--> 184                               (row_stride, col_stride), padding_type)
    185 
    186 

/home/rustam/anaconda2/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.pyc in get_conv_output_size(input_size, filter_size, strides, padding_type)
    147          zip(filter_size, input_size)):
    148     raise ValueError("Filter must not be larger than the input: "
--> 149                      "Filter: %r Input: %r" % (filter_size, input_size))
    150 
    151   if padding_type == b"VALID":

ValueError: Filter must not be larger than the input: Filter: (3, 3) Input: (150, 2)

[rykov8]

Wow, this error looks really super strange. Are you sure, that you haven't change something in SSD function accidentally?
Also, as I see, the error refers to _/opt/neural/ssd_keras/ssd.pyc,_ I am afraid, that .pyc file could be obsolete and can have some Theano backend footprints. Try do delete ssd.pyc (NOT ssd.py) and, please, comment, if it helps.

[fortunto2]

no... not helped delete all .pyc

[rykov8]

@fortunto2 in your ~/.keras/keras.json do you have both "image_dim_ordering": "tf" and "backend": "tensorflow" like in Keras docs?

[fortunto2]

my keras 1.1.1 py27_0, tensorflow 0.10.0rc0 np111py27_0

[fortunto2]

yes! thanks... not switch image_dim_ordering

[rykov8]

Hooray! We've done it =)

[fortunto2]

all works! closed
thanks