Skip to content

Fix RoiAlign aligned=True export #4692

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Oct 26, 2021
Merged

Fix RoiAlign aligned=True export #4692

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2c11506
RoiAlign aligned=True
jiafatom Oct 21, 2021
1676c2e
Merge branch 'main' into roi_true
prabhat00155 Oct 26, 2021
e963b29
Merge branch 'main' into roi_true
prabhat00155 Oct 26, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 5 additions & 4 deletions torchvision/ops/_register_onnx_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,13 +26,14 @@ def roi_align(g, input, rois, spatial_scale, pooled_height, pooled_width, sampli
g, squeeze(g, select(g, rois, 1, g.op("Constant", value_t=torch.tensor([0], dtype=torch.long))), 1), False
)
rois = select(g, rois, 1, g.op("Constant", value_t=torch.tensor([1, 2, 3, 4], dtype=torch.long)))
# TODO: Remove this warning after ONNX opset 16 is supported.
if aligned:
warnings.warn(
"ONNX export of ROIAlign with aligned=True does not match PyTorch when using malformed boxes,"
" ONNX forces ROIs to be 1x1 or larger."
"ROIAlign with aligned=True is not supported in ONNX, but will be supported in opset 16. "
"The workaround is that the user need apply the patch "
"https://github.com/microsoft/onnxruntime/pull/8564 "
"and build ONNXRuntime from source."
)
scale = torch.tensor(0.5 / spatial_scale).to(dtype=torch.float)
rois = g.op("Sub", rois, scale)

# ONNX doesn't support negative sampling_ratio
if sampling_ratio < 0:
Expand Down