2

我在我的 Tensorflow 模型中使用了 Tensorflow 操作 ResizeArea、Select、Fill 和 Equal。当模型转换为 uff 时，收到警告为

Warning: No conversion function registered for layer: ResizeArea yet.
Converting upsample_heatmat as custom op: ResizeArea
Warning: No conversion function registered for layer: Select yet.
Converting Select as custom op: Select
Warning: No conversion function registered for layer: Fill yet.
Converting zeros_like as custom op: Fill
Warning: No conversion function registered for layer: Equal yet.
Converting Equal as custom op: Equal

因此，为 ResizeArea、Select、Fill 和 Equal 创建了插件。

然后插件被映射到 Tensorflow 操作为

class ModelData(object):
    INPUT_NAME = "image"
    EQUAL_NAME = "Equal"
    SELECT_NAME = "Select"
    PMAT_NAME = "upsample_pafmat"
    ZERO_LIKE = "zeros_like"
    HMAT_NAME = "upsample_heatmat"
    OUTPUT_NAME = "Openpose/output"

def prepare_namespace_plugin_map():
    # In this sample, the only operation that is not supported by TensorRT
    # is tf.nn.relu6, so we create a new node which will tell UffParser which
    # plugin to run and with which arguments in place of tf.nn.relu6.


    # The "clipMin" and "clipMax" fields of this TensorFlow node will be parsed by createPlugin,
    # and used to create a CustomClipPlugin with the appropriate parameters.
    trt_resizearea = gs.create_plugin_node(name="trt_resizearea", op="ResizeAreaPlugin", in_width=80.0, in_height=60.0, in_channel=3.0, upscale=4.0)
    trt_fill = gs.create_plugin_node(name="trt_fill", op="FillPlugin", in_width=320.0, in_height=240.0, in_channel=3.0, value=0.0)#fill 0
    trt_equal = gs.create_plugin_node(name="trt_equal", op="EqualPlugin", in_width=320.0, in_height=240.0, in_channel=3.0)
    trt_select = gs.create_plugin_node(name="trt_select", op="SelectPlugin", in_width=320.0, in_height=240.0, value=0.0)
    namespace_plugin_map = {
        ModelData.SELECT_NAME: trt_select,
        ModelData.EQUAL_NAME: trt_equal,
        ModelData.PMAT_NAME: trt_resizearea,
        ModelData.HMAT_NAME: trt_resizearea,
        ModelData.ZERO_LIKE: trt_fill
    }
    return namespace_plugin_map

def model_to_uff(model_path):
    # Transform graph using graphsurgeon to map unsupported TensorFlow
    # operations to appropriate TensorRT custom layer plugins
    dynamic_graph = gs.DynamicGraph(model_path)
    dynamic_graph.collapse_namespaces(prepare_namespace_plugin_map())
    # Save resulting graph to UFF file
    output_uff_path = model_path_to_uff_path(model_path)
    uff.from_tensorflow(
        dynamic_graph.as_graph_def(),
        [ModelData.OUTPUT_NAME],
        output_filename=output_uff_path,
        text=True
    )
    return output_uff_path

def model_path_to_uff_path(model_path):
    uff_path = os.path.splitext(model_path)[0] + ".uff"
    return uff_path

为什么我仍然有警告

Warning: No conversion function registered for layer: ResizeAreaPlugin yet.
Converting trt_resizearea as custom op: ResizeAreaPlugin
W0808 17:44:51.442725 139793630279424 deprecation_wrapper.py:119] From /home/coie/Data/coie/Softwares/venv/lib/python3.5/site-packages/uff/converters/tensorflow/converter.py:179: The name tf.AttrValue is deprecated. Please use tf.compat.v1.AttrValue instead.

Warning: No conversion function registered for layer: SelectPlugin yet.
Converting trt_select as custom op: SelectPlugin
Warning: No conversion function registered for layer: FillPlugin yet.
Converting trt_fill as custom op: FillPlugin
Warning: No conversion function registered for layer: EqualPlugin yet.
Converting trt_equal as custom op: EqualPlugin

有什么问题？

4

1 回答 1

0

我的解决方案是正确的。尽管插件已成功映射到 Tensorflow 操作，但仍会产生警告。我们可以通过在插件 API 中打印来检查插件是否在运行的 TensorRT 引擎中成功加载。

于 2019-08-16T02:19:21.290 回答