Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #20904 from Crayon-new:fix_bug_in_maxLayer

Browse Source 
fix bug: wrong output dimension when "keep_dims" is false in pooling layer.

* fix bug in max layer

* code align

* delete permute layer and add test case

* add name assert

* check other cases

* remove c++11 features

* style:add "const" remove assert

* style:sanitize file names
pull/21030/head
ZaKiiiiiiiii 3 years ago committed by GitHub
parent 1ac7baceff
commit 98b6ce353c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 66 additions and 24 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 77
      modules/dnn/src/tensorflow/tf_importer.cpp
  2. 13
      modules/dnn/test/test_tf_importer.cpp

77
modules/dnn/src/tensorflow/tf_importer.cpp
Unescape View File

@ -2142,6 +2142,7 @@ void TFImporter::parseMean(tensorflow::GraphDef& net, const tensorflow::NodeDef&
const std::string& type = layer.op();
const int num_inputs = layer.input_size();
std::string pool_type = cv::toLowerCase(type);
DataLayout layout = getDataLayout(name, data_layouts);
if (pool_type == "mean")
{
@ -2205,6 +2206,16 @@ void TFImporter::parseMean(tensorflow::GraphDef& net, const tensorflow::NodeDef&
if (!keepDims)
{
if (layout == DATA_LAYOUT_NHWC)
{
LayerParams permLP;
int order[] = {0, 2, 3, 1}; // From OpenCV's NCHW to NHWC.
std::string permName = name + "/nhwc";
Pin inpId = Pin(layerShapeName);
addPermuteLayer(order, permName, inpId);
layerShapeName = permName;
}
LayerParams squeezeLp;
std::string squeezeName = name + "/squeeze";
CV_Assert(layer_id.find(squeezeName) == layer_id.end());
@ -2227,22 +2238,30 @@ void TFImporter::parseMean(tensorflow::GraphDef& net, const tensorflow::NodeDef&
layerParams.set("pool", pool_type);
layerParams.set(axis == 2 ? "kernel_w" : "kernel_h", 1);
layerParams.set(axis == 2 ? "global_pooling_h" : "global_pooling_w", true);
int id = dstNet.addLayer(name, "Pooling", layerParams);
layer_id[name] = id;
connect(layer_id, dstNet, parsePin(layer.input(0)), id, 0);
if (!keepDims)
if (keepDims)
{
int id = dstNet.addLayer(name, "Pooling", layerParams);
layer_id[name] = id;
connect(layer_id, dstNet, parsePin(layer.input(0)), id, 0);
}
else
{
// To keep correct order after squeeze dims we first need to change layout from NCHW to NHWC
std::string poolingName = name + "/Pooling";
CV_Assert(layer_id.find(poolingName) == layer_id.end());
int id = dstNet.addLayer(poolingName, "Pooling", layerParams);
layer_id[poolingName] = id;
connect(layer_id, dstNet, parsePin(layer.input(0)), id, 0);
LayerParams permLP;
int order[] = {0, 2, 3, 1}; // From OpenCV's NCHW to NHWC.
std::string permName = name + "/nchw";
Pin inpId = Pin(name);
std::string permName = name + "/nhwc";
Pin inpId = Pin(poolingName);
addPermuteLayer(order, permName, inpId);
LayerParams squeezeLp;
std::string squeezeName = name + "/squeeze";
CV_Assert(layer_id.find(squeezeName) == layer_id.end());
const std::string& squeezeName = name;
squeezeLp.set("axis", indices.at<int>(0));
squeezeLp.set("end_axis", indices.at<int>(0) + 1);
int squeezeId = dstNet.addLayer(squeezeName, "Flatten", squeezeLp);
@ -2254,32 +2273,34 @@ void TFImporter::parseMean(tensorflow::GraphDef& net, const tensorflow::NodeDef&
{
int order[] = {0, 2, 3, 1}; // From OpenCV's NCHW to NHWC.
Pin inpId = parsePin(layer.input(0));
addPermuteLayer(order, name + "/nhwc", inpId);
std::string permName = name + "/nhwc";
addPermuteLayer(order, permName, inpId);
layerParams.set("pool", pool_type);
layerParams.set("kernel_h", 1);
layerParams.set("global_pooling_w", true);
int id = dstNet.addLayer(name, "Pooling", layerParams);
layer_id[name] = id;
connect(layer_id, dstNet, inpId, id, 0);
std::string poolingName = name + "/Pooling";
CV_Assert(layer_id.find(poolingName) == layer_id.end());
int id = dstNet.addLayer(poolingName, "Pooling", layerParams);
layer_id[poolingName] = id;
connect(layer_id, dstNet, Pin(permName), id, 0);
if (!keepDims)
{
LayerParams squeezeLp;
std::string squeezeName = name + "/squeeze";
CV_Assert(layer_id.find(squeezeName) == layer_id.end());
const std::string& squeezeName = name;
int channel_id = 3; // TF NHWC layout
squeezeLp.set("axis", channel_id - 1);
squeezeLp.set("end_axis", channel_id);
int squeezeId = dstNet.addLayer(squeezeName, "Flatten", squeezeLp);
layer_id[squeezeName] = squeezeId;
connect(layer_id, dstNet, Pin(name), squeezeId, 0);
connect(layer_id, dstNet, Pin(poolingName), squeezeId, 0);
}
else
{
int order[] = {0, 3, 1, 2}; // From NHWC to OpenCV's NCHW.
Pin inpId = parsePin(name);
addPermuteLayer(order, name + "/nchw", inpId);
Pin inpId = parsePin(poolingName);
addPermuteLayer(order, name, inpId);
}
}
} else {
@ -2288,18 +2309,26 @@ void TFImporter::parseMean(tensorflow::GraphDef& net, const tensorflow::NodeDef&
layerParams.set("pool", pool_type);
layerParams.set("global_pooling", true);
int id = dstNet.addLayer(name, "Pooling", layerParams);
layer_id[name] = id;
connect(layer_id, dstNet, parsePin(layer.input(0)), id, 0);
if (!keepDims)
if (keepDims)
{
int id = dstNet.addLayer(name, "Pooling", layerParams);
layer_id[name] = id;
connect(layer_id, dstNet, parsePin(layer.input(0)), id, 0);
}
else
{
std::string poolingName = name + "/Pooling";
CV_Assert(layer_id.find(poolingName) == layer_id.end());
int id = dstNet.addLayer(poolingName, "Pooling", layerParams);
layer_id[poolingName] = id;
connect(layer_id, dstNet, parsePin(layer.input(0)), id, 0);
LayerParams flattenLp;
std::string flattenName = name + "/flatten";
CV_Assert(layer_id.find(flattenName) == layer_id.end());
const std::string& flattenName = name;
int flattenId = dstNet.addLayer(flattenName, "Flatten", flattenLp);
layer_id[flattenName] = flattenId;
connect(layer_id, dstNet, Pin(name), flattenId, 0);
connect(layer_id, dstNet, Pin(poolingName), flattenId, 0);
data_layouts[name] = DATA_LAYOUT_PLANAR;
}
}
}

13
modules/dnn/test/test_tf_importer.cpp
Unescape View File

@ -413,6 +413,19 @@ TEST_P(Test_TensorFlow_layers, pooling_reduce_sum)
runTensorFlowNet("reduce_sum"); // a SUM pooling over all spatial dimensions.
}
TEST_P(Test_TensorFlow_layers, pooling_reduce_sum2)
{
int axises[] = {0, 1, 2, 3};
for (int keepdims = 0; keepdims <= 1; ++keepdims)
{
for (int i = 0; i < sizeof(axises)/sizeof(axises[0]); ++i)
{
runTensorFlowNet(cv::format("reduce_sum_%d_%s", axises[i], (keepdims ? "True" : "False")));
}
runTensorFlowNet(cv::format("reduce_sum_1_2_%s", keepdims ? "True" : "False"));
}
}
TEST_P(Test_TensorFlow_layers, max_pool_grad)
{
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)

Write Preview
Loading…
Cancel
Save