search

serizba / cppflow

Run TensorFlow models in C++ without installation and without Bazel
https://serizba.github.io/cppflow/
MIT License
785 stars 178 forks source link

Default MaxPoolingOp only supports NHWC on device type CPU but I don't have MaxPooling in this model #262

Open DemO-O-On opened 4 months ago

DemO-O-On commented 4 months ago

Hello, I want to make prediction from this:

auto input = cppflow::tensor(points_teeth, shape);
for (auto s: model_points.get_operations()){
      std::cout << s << std::endl;
}
output = model_points({{"serving_default_input_3:0", input}},{"StatefulPartitionedCall:0"});

Here's what this code prints:

count/Read/ReadVariableOp
total
total/Read/ReadVariableOp
count_1
count_1/Read/ReadVariableOp
total_1
total_1/Read/ReadVariableOp
count_2
count_2/Read/ReadVariableOp
total_2
total_2/Read/ReadVariableOp
Adam/v/conv1d_20/bias
Adam/v/conv1d_20/bias/Read/ReadVariableOp
Adam/m/conv1d_20/bias
Adam/m/conv1d_20/bias/Read/ReadVariableOp
Adam/v/conv1d_20/kernel
Adam/v/conv1d_20/kernel/Read/ReadVariableOp
Adam/m/conv1d_20/kernel
Adam/m/conv1d_20/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_35/beta
Adam/v/batch_normalization_35/beta/Read/ReadVariableOp
Adam/m/batch_normalization_35/beta
Adam/m/batch_normalization_35/beta/Read/ReadVariableOp
Adam/v/batch_normalization_35/gamma
Adam/v/batch_normalization_35/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_35/gamma
Adam/m/batch_normalization_35/gamma/Read/ReadVariableOp
Adam/v/conv1d_19/bias
Adam/v/conv1d_19/bias/Read/ReadVariableOp
Adam/m/conv1d_19/bias
Adam/m/conv1d_19/bias/Read/ReadVariableOp
Adam/v/conv1d_19/kernel
Adam/v/conv1d_19/kernel/Read/ReadVariableOp
Adam/m/conv1d_19/kernel
Adam/m/conv1d_19/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_34/beta
Adam/v/batch_normalization_34/beta/Read/ReadVariableOp
Adam/m/batch_normalization_34/beta
Adam/m/batch_normalization_34/beta/Read/ReadVariableOp
Adam/v/batch_normalization_34/gamma
Adam/v/batch_normalization_34/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_34/gamma
Adam/m/batch_normalization_34/gamma/Read/ReadVariableOp
Adam/v/conv1d_transpose_17/bias
Adam/v/conv1d_transpose_17/bias/Read/ReadVariableOp
Adam/m/conv1d_transpose_17/bias
Adam/m/conv1d_transpose_17/bias/Read/ReadVariableOp
Adam/v/conv1d_transpose_17/kernel
Adam/v/conv1d_transpose_17/kernel/Read/ReadVariableOp
Adam/m/conv1d_transpose_17/kernel
Adam/m/conv1d_transpose_17/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_33/beta
Adam/v/batch_normalization_33/beta/Read/ReadVariableOp
Adam/m/batch_normalization_33/beta
Adam/m/batch_normalization_33/beta/Read/ReadVariableOp
Adam/v/batch_normalization_33/gamma
Adam/v/batch_normalization_33/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_33/gamma
Adam/m/batch_normalization_33/gamma/Read/ReadVariableOp
Adam/v/conv1d_transpose_16/bias
Adam/v/conv1d_transpose_16/bias/Read/ReadVariableOp
Adam/m/conv1d_transpose_16/bias
Adam/m/conv1d_transpose_16/bias/Read/ReadVariableOp
Adam/v/conv1d_transpose_16/kernel
Adam/v/conv1d_transpose_16/kernel/Read/ReadVariableOp
Adam/m/conv1d_transpose_16/kernel
Adam/m/conv1d_transpose_16/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_32/beta
Adam/v/batch_normalization_32/beta/Read/ReadVariableOp
Adam/m/batch_normalization_32/beta
Adam/m/batch_normalization_32/beta/Read/ReadVariableOp
Adam/v/batch_normalization_32/gamma
Adam/v/batch_normalization_32/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_32/gamma
Adam/m/batch_normalization_32/gamma/Read/ReadVariableOp
Adam/v/conv1d_transpose_15/bias
Adam/v/conv1d_transpose_15/bias/Read/ReadVariableOp
Adam/m/conv1d_transpose_15/bias
Adam/m/conv1d_transpose_15/bias/Read/ReadVariableOp
Adam/v/conv1d_transpose_15/kernel
Adam/v/conv1d_transpose_15/kernel/Read/ReadVariableOp
Adam/m/conv1d_transpose_15/kernel
Adam/m/conv1d_transpose_15/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_31/beta
Adam/v/batch_normalization_31/beta/Read/ReadVariableOp
Adam/m/batch_normalization_31/beta
Adam/m/batch_normalization_31/beta/Read/ReadVariableOp
Adam/v/batch_normalization_31/gamma
Adam/v/batch_normalization_31/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_31/gamma
Adam/m/batch_normalization_31/gamma/Read/ReadVariableOp
Adam/v/conv1d_transpose_14/bias
Adam/v/conv1d_transpose_14/bias/Read/ReadVariableOp
Adam/m/conv1d_transpose_14/bias
Adam/m/conv1d_transpose_14/bias/Read/ReadVariableOp
Adam/v/conv1d_transpose_14/kernel
Adam/v/conv1d_transpose_14/kernel/Read/ReadVariableOp
Adam/m/conv1d_transpose_14/kernel
Adam/m/conv1d_transpose_14/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_30/beta
Adam/v/batch_normalization_30/beta/Read/ReadVariableOp
Adam/m/batch_normalization_30/beta
Adam/m/batch_normalization_30/beta/Read/ReadVariableOp
Adam/v/batch_normalization_30/gamma
Adam/v/batch_normalization_30/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_30/gamma
Adam/m/batch_normalization_30/gamma/Read/ReadVariableOp
Adam/v/conv1d_transpose_13/bias
Adam/v/conv1d_transpose_13/bias/Read/ReadVariableOp
Adam/m/conv1d_transpose_13/bias
Adam/m/conv1d_transpose_13/bias/Read/ReadVariableOp
Adam/v/conv1d_transpose_13/kernel
Adam/v/conv1d_transpose_13/kernel/Read/ReadVariableOp
Adam/m/conv1d_transpose_13/kernel
Adam/m/conv1d_transpose_13/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_29/beta
Adam/v/batch_normalization_29/beta/Read/ReadVariableOp
Adam/m/batch_normalization_29/beta
Adam/m/batch_normalization_29/beta/Read/ReadVariableOp
Adam/v/batch_normalization_29/gamma
Adam/v/batch_normalization_29/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_29/gamma
Adam/m/batch_normalization_29/gamma/Read/ReadVariableOp
Adam/v/conv1d_transpose_12/bias
Adam/v/conv1d_transpose_12/bias/Read/ReadVariableOp
Adam/m/conv1d_transpose_12/bias
Adam/m/conv1d_transpose_12/bias/Read/ReadVariableOp
Adam/v/conv1d_transpose_12/kernel
Adam/v/conv1d_transpose_12/kernel/Read/ReadVariableOp
Adam/m/conv1d_transpose_12/kernel
Adam/m/conv1d_transpose_12/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_28/beta
Adam/v/batch_normalization_28/beta/Read/ReadVariableOp
Adam/m/batch_normalization_28/beta
Adam/m/batch_normalization_28/beta/Read/ReadVariableOp
Adam/v/batch_normalization_28/gamma
Adam/v/batch_normalization_28/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_28/gamma
Adam/m/batch_normalization_28/gamma/Read/ReadVariableOp
Adam/v/conv1d_18/bias
Adam/v/conv1d_18/bias/Read/ReadVariableOp
Adam/m/conv1d_18/bias
Adam/m/conv1d_18/bias/Read/ReadVariableOp
Adam/v/conv1d_18/kernel
Adam/v/conv1d_18/kernel/Read/ReadVariableOp
Adam/m/conv1d_18/kernel
Adam/m/conv1d_18/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_27/beta
Adam/v/batch_normalization_27/beta/Read/ReadVariableOp
Adam/m/batch_normalization_27/beta
Adam/m/batch_normalization_27/beta/Read/ReadVariableOp
Adam/v/batch_normalization_27/gamma
Adam/v/batch_normalization_27/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_27/gamma
Adam/m/batch_normalization_27/gamma/Read/ReadVariableOp
Adam/v/conv1d_17/bias
Adam/v/conv1d_17/bias/Read/ReadVariableOp
Adam/m/conv1d_17/bias
Adam/m/conv1d_17/bias/Read/ReadVariableOp
Adam/v/conv1d_17/kernel
Adam/v/conv1d_17/kernel/Read/ReadVariableOp
Adam/m/conv1d_17/kernel
Adam/m/conv1d_17/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_26/beta
Adam/v/batch_normalization_26/beta/Read/ReadVariableOp
Adam/m/batch_normalization_26/beta
Adam/m/batch_normalization_26/beta/Read/ReadVariableOp
Adam/v/batch_normalization_26/gamma
Adam/v/batch_normalization_26/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_26/gamma
Adam/m/batch_normalization_26/gamma/Read/ReadVariableOp
Adam/v/conv1d_16/bias
Adam/v/conv1d_16/bias/Read/ReadVariableOp
Adam/m/conv1d_16/bias
Adam/m/conv1d_16/bias/Read/ReadVariableOp
Adam/v/conv1d_16/kernel
Adam/v/conv1d_16/kernel/Read/ReadVariableOp
Adam/m/conv1d_16/kernel
Adam/m/conv1d_16/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_25/beta
Adam/v/batch_normalization_25/beta/Read/ReadVariableOp
Adam/m/batch_normalization_25/beta
Adam/m/batch_normalization_25/beta/Read/ReadVariableOp
Adam/v/batch_normalization_25/gamma
Adam/v/batch_normalization_25/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_25/gamma
Adam/m/batch_normalization_25/gamma/Read/ReadVariableOp
Adam/v/conv1d_15/bias
Adam/v/conv1d_15/bias/Read/ReadVariableOp
Adam/m/conv1d_15/bias
Adam/m/conv1d_15/bias/Read/ReadVariableOp
Adam/v/conv1d_15/kernel
Adam/v/conv1d_15/kernel/Read/ReadVariableOp
Adam/m/conv1d_15/kernel
Adam/m/conv1d_15/kernel/Read/ReadVariableOp
Adam/v/batch_normalization_24/beta
Adam/v/batch_normalization_24/beta/Read/ReadVariableOp
Adam/m/batch_normalization_24/beta
Adam/m/batch_normalization_24/beta/Read/ReadVariableOp
Adam/v/batch_normalization_24/gamma
Adam/v/batch_normalization_24/gamma/Read/ReadVariableOp
Adam/m/batch_normalization_24/gamma
Adam/m/batch_normalization_24/gamma/Read/ReadVariableOp
Adam/v/conv1d_14/bias
Adam/v/conv1d_14/bias/Read/ReadVariableOp
Adam/m/conv1d_14/bias
Adam/m/conv1d_14/bias/Read/ReadVariableOp
Adam/v/conv1d_14/kernel
Adam/v/conv1d_14/kernel/Read/ReadVariableOp
Adam/m/conv1d_14/kernel
Adam/m/conv1d_14/kernel/Read/ReadVariableOp
learning_rate
learning_rate/Read/ReadVariableOp
iteration
iteration/Read/ReadVariableOp
conv1d_20/bias
conv1d_20/bias/Read/ReadVariableOp
conv1d_20/kernel
conv1d_20/kernel/Read/ReadVariableOp
batch_normalization_35/moving_variance
batch_normalization_35/moving_variance/Read/ReadVariableOp
batch_normalization_35/moving_mean
batch_normalization_35/moving_mean/Read/ReadVariableOp
batch_normalization_35/beta
batch_normalization_35/beta/Read/ReadVariableOp
batch_normalization_35/gamma
batch_normalization_35/gamma/Read/ReadVariableOp
conv1d_19/bias
conv1d_19/bias/Read/ReadVariableOp
conv1d_19/kernel
conv1d_19/kernel/Read/ReadVariableOp
batch_normalization_34/moving_variance
batch_normalization_34/moving_variance/Read/ReadVariableOp
batch_normalization_34/moving_mean
batch_normalization_34/moving_mean/Read/ReadVariableOp
batch_normalization_34/beta
batch_normalization_34/beta/Read/ReadVariableOp
batch_normalization_34/gamma
batch_normalization_34/gamma/Read/ReadVariableOp
conv1d_transpose_17/bias
conv1d_transpose_17/bias/Read/ReadVariableOp
conv1d_transpose_17/kernel
conv1d_transpose_17/kernel/Read/ReadVariableOp
batch_normalization_33/moving_variance
batch_normalization_33/moving_variance/Read/ReadVariableOp
batch_normalization_33/moving_mean
batch_normalization_33/moving_mean/Read/ReadVariableOp
batch_normalization_33/beta
batch_normalization_33/beta/Read/ReadVariableOp
batch_normalization_33/gamma
batch_normalization_33/gamma/Read/ReadVariableOp
conv1d_transpose_16/bias
conv1d_transpose_16/bias/Read/ReadVariableOp
conv1d_transpose_16/kernel
conv1d_transpose_16/kernel/Read/ReadVariableOp
batch_normalization_32/moving_variance
batch_normalization_32/moving_variance/Read/ReadVariableOp
batch_normalization_32/moving_mean
batch_normalization_32/moving_mean/Read/ReadVariableOp
batch_normalization_32/beta
batch_normalization_32/beta/Read/ReadVariableOp
batch_normalization_32/gamma
batch_normalization_32/gamma/Read/ReadVariableOp
conv1d_transpose_15/bias
conv1d_transpose_15/bias/Read/ReadVariableOp
conv1d_transpose_15/kernel
conv1d_transpose_15/kernel/Read/ReadVariableOp
batch_normalization_31/moving_variance
batch_normalization_31/moving_variance/Read/ReadVariableOp
batch_normalization_31/moving_mean
batch_normalization_31/moving_mean/Read/ReadVariableOp
batch_normalization_31/beta
batch_normalization_31/beta/Read/ReadVariableOp
batch_normalization_31/gamma
batch_normalization_31/gamma/Read/ReadVariableOp
conv1d_transpose_14/bias
conv1d_transpose_14/bias/Read/ReadVariableOp
conv1d_transpose_14/kernel
conv1d_transpose_14/kernel/Read/ReadVariableOp
batch_normalization_30/moving_variance
batch_normalization_30/moving_variance/Read/ReadVariableOp
batch_normalization_30/moving_mean
batch_normalization_30/moving_mean/Read/ReadVariableOp
batch_normalization_30/beta
batch_normalization_30/beta/Read/ReadVariableOp
batch_normalization_30/gamma
batch_normalization_30/gamma/Read/ReadVariableOp
conv1d_transpose_13/bias
conv1d_transpose_13/bias/Read/ReadVariableOp
conv1d_transpose_13/kernel
conv1d_transpose_13/kernel/Read/ReadVariableOp
batch_normalization_29/moving_variance
batch_normalization_29/moving_variance/Read/ReadVariableOp
batch_normalization_29/moving_mean
batch_normalization_29/moving_mean/Read/ReadVariableOp
batch_normalization_29/beta
batch_normalization_29/beta/Read/ReadVariableOp
batch_normalization_29/gamma
batch_normalization_29/gamma/Read/ReadVariableOp
conv1d_transpose_12/bias
conv1d_transpose_12/bias/Read/ReadVariableOp
conv1d_transpose_12/kernel
conv1d_transpose_12/kernel/Read/ReadVariableOp
batch_normalization_28/moving_variance
batch_normalization_28/moving_variance/Read/ReadVariableOp
batch_normalization_28/moving_mean
batch_normalization_28/moving_mean/Read/ReadVariableOp
batch_normalization_28/beta
batch_normalization_28/beta/Read/ReadVariableOp
batch_normalization_28/gamma
batch_normalization_28/gamma/Read/ReadVariableOp
conv1d_18/bias
conv1d_18/bias/Read/ReadVariableOp
conv1d_18/kernel
conv1d_18/kernel/Read/ReadVariableOp
batch_normalization_27/moving_variance
batch_normalization_27/moving_variance/Read/ReadVariableOp
batch_normalization_27/moving_mean
batch_normalization_27/moving_mean/Read/ReadVariableOp
batch_normalization_27/beta
batch_normalization_27/beta/Read/ReadVariableOp
batch_normalization_27/gamma
batch_normalization_27/gamma/Read/ReadVariableOp
conv1d_17/bias
conv1d_17/bias/Read/ReadVariableOp
conv1d_17/kernel
conv1d_17/kernel/Read/ReadVariableOp
batch_normalization_26/moving_variance
batch_normalization_26/moving_variance/Read/ReadVariableOp
batch_normalization_26/moving_mean
batch_normalization_26/moving_mean/Read/ReadVariableOp
batch_normalization_26/beta
batch_normalization_26/beta/Read/ReadVariableOp
batch_normalization_26/gamma
batch_normalization_26/gamma/Read/ReadVariableOp
conv1d_16/bias
conv1d_16/bias/Read/ReadVariableOp
conv1d_16/kernel
conv1d_16/kernel/Read/ReadVariableOp
batch_normalization_25/moving_variance
batch_normalization_25/moving_variance/Read/ReadVariableOp
batch_normalization_25/moving_mean
batch_normalization_25/moving_mean/Read/ReadVariableOp
batch_normalization_25/beta
batch_normalization_25/beta/Read/ReadVariableOp
batch_normalization_25/gamma
batch_normalization_25/gamma/Read/ReadVariableOp
conv1d_15/bias
conv1d_15/bias/Read/ReadVariableOp
conv1d_15/kernel
conv1d_15/kernel/Read/ReadVariableOp
batch_normalization_24/moving_variance
batch_normalization_24/moving_variance/Read/ReadVariableOp
batch_normalization_24/moving_mean
batch_normalization_24/moving_mean/Read/ReadVariableOp
batch_normalization_24/beta
batch_normalization_24/beta/Read/ReadVariableOp
batch_normalization_24/gamma
batch_normalization_24/gamma/Read/ReadVariableOp
conv1d_14/bias
conv1d_14/bias/Read/ReadVariableOp
conv1d_14/kernel
conv1d_14/kernel/Read/ReadVariableOp
serving_default_input_3
StatefulPartitionedCall
NoOp
Const
saver_filename
StatefulPartitionedCall_1
StatefulPartitionedCall_2
terminate called after throwing an instance of 'std::runtime_error'
  what():  Default MaxPoolingOp only supports NHWC on device type CPU
         [[{{function_node __inference__wrapped_model_1115831}}{{node model_2/max_pooling1d_2/MaxPool}}]]

I loaded another model in this file before and made predictions but I don't understand how and why they can interract. Also I noticed I have maxpooling in this model in python but don't have it in c++. Here's code of model in python:

input = Input(shape=(num_points, 3))
x1 = conv_block(input, 128)
x = conv_block(x1, 256)
x2 = conv_block(x, 1024)
x = conv_block(x2, 1024)
x3 = conv_block(x, 1024)
x7 = deconv_block(x3, 1024)
x = deconv_block(x7, 1024)
x4 = deconv_block(x, 1024)
x = deconv_block(x4, 512)
x5 = deconv_block(x, 256)
x = deconv_block(x5, 128)
x6 = MaxPool1D(4, data_format = 'channels_first')(x)
#x6 = tf.tile(x6, [1, num_points, 1])
segmentation_input = Concatenate()(
    [
        x1,
        x2,
        x4,
        x5,
        x6
    ]
)

segmentation_features = conv_block(segmentation_input, 256)

output = Conv1D(num_classes, 1, activation="sigmoid")(segmentation_features)
return Model(input, output)
DemO-O-On commented 4 months ago

Upd: resaved model in Google Colab using only CPU and tried to make prediction with it. Here's new error:

(tensor: shape=[3 32 3], dtype=TF_FLOAT, data=
[[[0.0123270601 0.355330735 0.547858953]
  [0.000193346394 0.366971523 0.547208309]
  [0.0271116234 0.350225359 0.546478629]
  ...
  [0.119589187 0.348327279 0.536324918]
  [0.0871538371 0.366260797 0.535787404]
  [0.0218100566 0.398851484 0.535461068]]

 [[0.173270345 0.320599109 0.53522557]
  [0.058475405 0.380493224 0.5350824]
  [0.221432179 0.282036275 0.535037518]
  ...
  [0.236433044 0.289447248 0.529477835]
  [0.720739186 0.13403447 0.52921474]
  [0.0364344493 0.41110763 0.528939784]]

 [[0.732707858 0.136695519 0.528835058]
  [0.227762714 0.299189806 0.528707266]
  [0.092518039 0.384792328 0.528624833]
  ...
  [0.665200353 0.13419053 0.524711549]
  [0.686733603 0.139694497 0.524705172]
  [0.156475887 0.3657493 0.524685383]]])
terminate called after throwing an instance of 'std::runtime_error'
  what():  Could not deduce type! type_name: St6vectorIN7cppflow6tensorESaIS1_EE

It appears after all operations inside of predict