Question

我正在尝试使用DeepLab转换由model zoo托管的tfjs convertor的冻结推断图。

转换成功，但是当我加载模型时，得到以下输出：

Error importing the graph: block_shape must be positive for 'MobilenetV2/expanded_conv_7/depthwise/depthwise/SpaceToBatchND' (op: 'SpaceToBatchND') with input shapes: [1,?,?,384], [2], [2,2] and with computed input tensors: input[1] = <0 0>, input[2] = <[0 0][0 0]>. Process finished with exit code 2

我尝试转换mobilenet冻结的图，然后加载它，成功了。我用于加载转换后的模型的脚本是：

graph_def = None
graph = None
model_file = "tensorflowjs_model.pb"

print('Loading graph definition ...', file=sys.stderr)
try:
    with tf.gfile.GFile(model_file, "rb") as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
except BaseException as e:
    parser.exit(2, 'Error loading the graph definition: {}'.format(str(e)))

print('Importing graph ...', file=sys.stderr)
try:
    assert graph_def is not None
    with tf.Graph().as_default() as graph:  # type: tf.Graph
        tf.import_graph_def(
            graph_def,
            name=''
        )
except BaseException as e:
    parser.exit(2, 'Error importing the graph: {}'.format(str(e)))

从TFJS转换程序加载转换后的Deeplab模型时出错

0 个答案: