mirrors
/
mmcv
mirror of https://github.com/open-mmlab/mmcv.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

[Fix] Force bbox_overlaps calculation with FP32 for ascend device (#2697) 

* modify bbox_overlaps op adapter

* update
pull/2700/head
liuhw 2023-03-24 19:08:57 +08:00 committed by GitHub
parent 5f1d55885e
commit 4c4ba6cb17
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 21 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
mmcv/ops/csrc/pytorch/npu/bbox_overlaps_npu.cpp
View File

@ -12,23 +12,33 @@ void bbox_overlaps_npu(const Tensor bboxes1, const Tensor bboxes2, Tensor ious,
if (mode == 1) { if (mode == 1) {
modeStr = "iof"; modeStr = "iof";
} }
float offset_ = 1; at::Tensor bboxesFP32 = bboxes2;
if (offset == 0) { at::Tensor gtboxesFP32 = bboxes1;
offset_ = 0.01; if (bboxes2.scalar_type() != at::ScalarType::Float) {
bboxesFP32 = NPUNativeFunctions::npu_dtype_cast(bboxes2, at::kFloat);
gtboxesFP32 = NPUNativeFunctions::npu_dtype_cast(bboxes1, at::kFloat);
} }
at::Tensor bboxes = at::ones_like(bboxes2); c10::SmallVector<int64_t, SIZE> iousSize = {gtboxesFP32.size(0),
at::Tensor gtboxes = at::ones_like(bboxes1); bboxesFP32.size(0)};
bboxes = aligned ? bboxes2.transpose(0, 1) : bboxes2; if (aligned) {
gtboxes = aligned ? bboxes1.transpose(0, 1) : bboxes1; iousSize = {gtboxesFP32.size(0), 1};
}
at::Tensor iousFP32 = OpPreparation::ApplyTensor(bboxesFP32, iousSize);
bboxesFP32 = aligned ? bboxesFP32.transpose(0, 1) : bboxesFP32;
gtboxesFP32 = aligned ? gtboxesFP32.transpose(0, 1) : gtboxesFP32;
OpCommand cmd; OpCommand cmd;
cmd.Name("Iou") cmd.Name("Iou")
.Input(bboxes) .Input(bboxesFP32)
.Input(gtboxes) .Input(gtboxesFP32)
.Output(ious) .Output(iousFP32)
.Attr("mode", modeStr) .Attr("mode", modeStr)
.Attr("eps", offset_) .Attr("eps", (float)offset)
.Attr("aligned", aligned) .Attr("aligned", aligned)
.Run(); .Run();
if (bboxes2.scalar_type() != at::ScalarType::Float) {
iousFP32 = NPUNativeFunctions::npu_dtype_cast(iousFP32, at::kHalf);
}
ious.copy_(iousFP32);
} }
REGISTER_NPU_IMPL(bbox_overlaps_impl, bbox_overlaps_npu); REGISTER_NPU_IMPL(bbox_overlaps_impl, bbox_overlaps_npu);