IntenseWebs/openvino
3
0 Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[LPT] fix weights output channel index in ConvolutionBackpropDataTransformation (#14124)

Browse Source 
ConvolutionBackpropData weights have following shape {input channels, output_channels, ...},
so unlike forward convolution, the output channels are on the second dimension.
This commit is contained in:
Mateusz Tabaka authored and GitHub committed 2022-11-23 09:01:59 +01:00
1 parent ae0b2ce3c3
commit e105b3ea37
4 files changed
+99 -29
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/common/low_precision_transformations/src/convolution_backprop_data.cpp
+1 -1
View File
@@ -149,7 +149,7 @@ bool ConvolutionBackpropDataTransformation::transform(TransformationContext &con
if (ov::is_type<opset1::FakeQuantize>(dequantization.data.get_node())) {
const std::shared_ptr<opset1::FakeQuantize> fq = ov::as_type_ptr<opset1::FakeQuantize>(dequantization.data.get_node_shared_ptr());
std::shared_ptr<ngraph::Node> newFQ = NetworkHelper::fold_fake_quantize(fq, true);
std::shared_ptr<ngraph::Node> newFQ = NetworkHelper::fold_fake_quantize(fq, true, 1);
NetworkHelper::copyInfo(fq, newFQ);
replace_node(fq, newFQ);
}