Skip to content

[ONNX] Fix dtype for NonMaxSuppression #7056

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 10 commits into from
Feb 15, 2023
Merged

[ONNX] Fix dtype for NonMaxSuppression #7056

Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
4949b5a
[ONNX] Fix dtype for NonMaxSuppression; misc improvements
justinchuby Dec 31, 2022
c088d46
Sort imports
justinchuby Dec 31, 2022
0a08a3c
tests
justinchuby Dec 31, 2022
0a63f3d
Merge branch 'main' into justinchuby/onnx
justinchuby Feb 13, 2023
45c8952
Merge branch 'main' into justinchuby/onnx
malfet Feb 14, 2023
e16575b
Merge branch 'main' into justinchuby/onnx
justinchuby Feb 14, 2023
5f5759a
Merge branch 'pytorch:main' into justinchuby/onnx
justinchuby Feb 14, 2023
a1b231a
Revert misc changes
justinchuby Feb 14, 2023
8e5a546
format
justinchuby Feb 15, 2023
8ae021b
Merge branch 'main' into justinchuby/onnx
pmeier Feb 15, 2023
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
14 changes: 9 additions & 5 deletions torchvision/ops/_register_onnx_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,21 +11,25 @@
def _register_custom_op():
from torch.onnx.symbolic_helper import parse_args
from torch.onnx.symbolic_opset11 import select, squeeze, unsqueeze
from torch.onnx.symbolic_opset9 import _cast_Long

@parse_args("v", "v", "f")
def symbolic_multi_label_nms(g, boxes, scores, iou_threshold):
boxes = unsqueeze(g, boxes, 0)
scores = unsqueeze(g, unsqueeze(g, scores, 0), 0)
max_output_per_class = g.op("Constant", value_t=torch.tensor([sys.maxsize], dtype=torch.long))
iou_threshold = g.op("Constant", value_t=torch.tensor([iou_threshold], dtype=torch.float))
nms_out = g.op("NonMaxSuppression", boxes, scores, max_output_per_class, iou_threshold)
nms_out = g.op(
"NonMaxSuppression",
g.op("Cast", boxes, to_i=torch.onnx.TensorProtoDataType.FLOAT),
g.op("Cast", scores, to_i=torch.onnx.TensorProtoDataType.FLOAT),
max_output_per_class,
iou_threshold,
)
return squeeze(g, select(g, nms_out, 1, g.op("Constant", value_t=torch.tensor([2], dtype=torch.long))), 1)

def _process_batch_indices_for_roi_align(g, rois):
return _cast_Long(
g, squeeze(g, select(g, rois, 1, g.op("Constant", value_t=torch.tensor([0], dtype=torch.long))), 1), False
)
indices = squeeze(g, select(g, rois, 1, g.op("Constant", value_t=torch.tensor([0], dtype=torch.long))), 1)
return g.op("Cast", indices, to_i=torch.onnx.TensorProtoDataType.INT64)

def _process_rois_for_roi_align(g, rois):
return select(g, rois, 1, g.op("Constant", value_t=torch.tensor([1, 2, 3, 4], dtype=torch.long)))
Expand Down