wehub-resource-sync
91 lines
3.2 KiB
C++
91 lines
3.2 KiB
C++
// Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#include "paddle/phi/kernels/yolo_box_kernel.h"
|
|
|
|
#include "paddle/phi/backends/xpu/enforce_xpu.h"
|
|
#include "paddle/phi/core/kernel_registry.h"
|
|
#include "paddle/phi/kernels/full_kernel.h"
|
|
#include "paddle/phi/kernels/funcs/yolo_box_util.h"
|
|
|
|
namespace phi {
|
|
|
|
template <typename T, typename Context>
|
|
void YoloBoxKernel(const Context& dev_ctx,
|
|
const DenseTensor& x,
|
|
const DenseTensor& img_size,
|
|
const std::vector<int>& anchors_,
|
|
int class_num,
|
|
float conf_thresh,
|
|
int downsample_ratio,
|
|
bool clip_bbox,
|
|
float scale_x_y,
|
|
bool iou_aware,
|
|
float iou_aware_factor,
|
|
DenseTensor* boxes,
|
|
DenseTensor* scores) {
|
|
if (x.numel() == 0 || img_size.numel() == 0) {
|
|
Full<T, Context>(dev_ctx, boxes->dims(), 0, boxes);
|
|
Full<T, Context>(dev_ctx, scores->dims(), 0, scores);
|
|
return;
|
|
}
|
|
|
|
using XPUType = typename XPUTypeTrait<T>::Type;
|
|
int r = 0;
|
|
auto* input = &x;
|
|
// auto* imgsize = &img_size;
|
|
float scale = scale_x_y;
|
|
float bias = -0.5f * (scale - 1.f);
|
|
|
|
std::vector<int64_t> anchors(anchors_.begin(), anchors_.end());
|
|
const int64_t n = input->dims()[0];
|
|
const int64_t h = input->dims()[2];
|
|
const int64_t w = input->dims()[3];
|
|
const int64_t box_num = boxes->dims()[1];
|
|
const int64_t an_num = anchors.size() / 2;
|
|
|
|
boxes->Resize({n, box_num, 4});
|
|
dev_ctx.template Alloc<T>(boxes);
|
|
|
|
scores->Resize({n, box_num, class_num});
|
|
dev_ctx.template Alloc<T>(scores);
|
|
|
|
auto x_data = reinterpret_cast<const XPUType*>(x.data<T>());
|
|
auto img_size_data = reinterpret_cast<const int*>(img_size.data<int>());
|
|
auto boxes_data = reinterpret_cast<XPUType*>(boxes->data<T>());
|
|
auto scores_data = reinterpret_cast<XPUType*>(scores->data<T>());
|
|
|
|
r = xpu::yolo_box<float>(dev_ctx.x_context(),
|
|
x_data,
|
|
img_size_data,
|
|
boxes_data,
|
|
scores_data,
|
|
n,
|
|
h,
|
|
w,
|
|
anchors,
|
|
an_num,
|
|
class_num,
|
|
conf_thresh,
|
|
downsample_ratio,
|
|
scale,
|
|
bias,
|
|
false);
|
|
PADDLE_ENFORCE_XDNN_SUCCESS(r, "yolo_box");
|
|
}
|
|
|
|
} // namespace phi
|
|
|
|
PD_REGISTER_KERNEL(yolo_box, XPU, ALL_LAYOUT, phi::YoloBoxKernel, float) {}
|