saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-23 22:13:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fixed a minor bug in layout transformation for Resize (#21954)

Browse source 
Since opset 18, 'scales' and 'sizes' constant inputs can be 2D tensors,
transpose for 2D tensors are not supported at current implementation,
fix it by only allowing 4D constant inputs.
This commit is contained in:
Wanming Lin 2024-11-01 05:20:10 +08:00 committed by GitHub
parent 55e0128b13
commit 7b9db658c3
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 5 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
onnxruntime/core/optimizer/layout_transformation/layout_transformation.cc
View file

@ -177,7 +177,11 @@ Status TransformLayoutForEP(Graph& graph, bool& modified, const IExecutionProvid
for (size_t i = 2; i < node->Inputs().size(); i++) {
auto constant = api_graph->GetConstant(node->Inputs()[i]);
if (constant != nullptr && constant->Data().size() > 0) {
input_perms.push_back(&input_perm);
// Starting from opset version 18, the 'scales' and 'sizes' can be any length up to the input rank.
// However, our current implementation only supports the transposition of 4D tensors.
if (constant->NumElements() == 4) {
input_perms.push_back(&input_perm);
}
} else {
// TODO: Fix inconsistency. We should Transpose the non-const inputs so that the result of our changes
// is consistent - all layout specific inputs are in NHWC format when we're done.