未验证 提交 fb93bd5c 编写于 作者: W wuyefeilin 提交者: GitHub

[phi] move yolov3_loss to phi (#40944)

* mv yolov3_loss op to phi

* fix as review

* update operator.h
上级 47383dca
...@@ -571,12 +571,17 @@ class OperatorWithKernel : public OperatorBase { ...@@ -571,12 +571,17 @@ class OperatorWithKernel : public OperatorBase {
if (has_phi_kernel) { if (has_phi_kernel) {
return true; return true;
} else { } else {
auto& op_kernels = OperatorWithKernel::AllOpKernels().at(type_); auto kernel_iter = OperatorWithKernel::AllOpKernels().find(type_);
return std::any_of( if (kernel_iter == OperatorWithKernel::AllOpKernels().end()) {
op_kernels.begin(), op_kernels.end(), return false;
[](OpKernelMap::const_reference kern_pair) { } else {
return platform::is_gpu_place(kern_pair.first.place_); auto& op_kernels = kernel_iter->second;
}); return std::any_of(
op_kernels.begin(), op_kernels.end(),
[](OpKernelMap::const_reference kern_pair) {
return platform::is_gpu_place(kern_pair.first.place_);
});
}
} }
} }
......
...@@ -9,10 +9,12 @@ ...@@ -9,10 +9,12 @@
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#include "paddle/fluid/operators/detection/yolov3_loss_op.h"
#include <memory> #include <memory>
#include "paddle/fluid/framework/infershape_utils.h"
#include "paddle/fluid/framework/op_registry.h" #include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/imperative/type_defs.h" #include "paddle/fluid/imperative/type_defs.h"
#include "paddle/phi/core/infermeta_utils.h"
#include "paddle/phi/infermeta/multiary.h"
namespace paddle { namespace paddle {
namespace operators { namespace operators {
...@@ -22,127 +24,6 @@ using framework::Tensor; ...@@ -22,127 +24,6 @@ using framework::Tensor;
class Yolov3LossOp : public framework::OperatorWithKernel { class Yolov3LossOp : public framework::OperatorWithKernel {
public: public:
using framework::OperatorWithKernel::OperatorWithKernel; using framework::OperatorWithKernel::OperatorWithKernel;
void InferShape(framework::InferShapeContext* ctx) const override {
OP_INOUT_CHECK(ctx->HasInput("X"), "Input", "X", "Yolov3LossOp");
OP_INOUT_CHECK(ctx->HasInput("GTBox"), "Input", "GTBox", "Yolov3LossOp");
OP_INOUT_CHECK(ctx->HasInput("GTLabel"), "Input", "GTLabel",
"Yolov3LossOp");
OP_INOUT_CHECK(ctx->HasOutput("Loss"), "Output", "Loss", "Yolov3LossOp");
OP_INOUT_CHECK(ctx->HasOutput("ObjectnessMask"), "Output", "ObjectnessMask",
"Yolov3LossOp");
OP_INOUT_CHECK(ctx->HasOutput("GTMatchMask"), "Output", "GTMatchMask",
"Yolov3LossOp");
auto dim_x = ctx->GetInputDim("X");
auto dim_gtbox = ctx->GetInputDim("GTBox");
auto dim_gtlabel = ctx->GetInputDim("GTLabel");
auto anchors = ctx->Attrs().Get<std::vector<int>>("anchors");
int anchor_num = anchors.size() / 2;
auto anchor_mask = ctx->Attrs().Get<std::vector<int>>("anchor_mask");
int mask_num = anchor_mask.size();
auto class_num = ctx->Attrs().Get<int>("class_num");
PADDLE_ENFORCE_EQ(dim_x.size(), 4,
platform::errors::InvalidArgument(
"Input(X) should be a 4-D tensor. But received "
"X dimension size(%s)",
dim_x.size()));
PADDLE_ENFORCE_EQ(dim_x[2], dim_x[3],
platform::errors::InvalidArgument(
"Input(X) dim[3] and dim[4] should be euqal."
"But received dim[3](%s) != dim[4](%s)",
dim_x[2], dim_x[3]));
PADDLE_ENFORCE_EQ(
dim_x[1], mask_num * (5 + class_num),
platform::errors::InvalidArgument(
"Input(X) dim[1] should be equal to (anchor_mask_number * (5 "
"+ class_num))."
"But received dim[1](%s) != (anchor_mask_number * "
"(5+class_num)(%s).",
dim_x[1], mask_num * (5 + class_num)));
PADDLE_ENFORCE_EQ(dim_gtbox.size(), 3,
platform::errors::InvalidArgument(
"Input(GTBox) should be a 3-D tensor, but "
"received gtbox dimension size(%s)",
dim_gtbox.size()));
PADDLE_ENFORCE_EQ(dim_gtbox[2], 4,
platform::errors::InvalidArgument(
"Input(GTBox) dim[2] should be 4",
"But receive dim[2](%s) != 5. ", dim_gtbox[2]));
PADDLE_ENFORCE_EQ(
dim_gtlabel.size(), 2,
platform::errors::InvalidArgument(
"Input(GTLabel) should be a 2-D tensor,"
"But received Input(GTLabel) dimension size(%s) != 2.",
dim_gtlabel.size()));
PADDLE_ENFORCE_EQ(
dim_gtlabel[0], dim_gtbox[0],
platform::errors::InvalidArgument(
"Input(GTBox) dim[0] and Input(GTLabel) dim[0] should be same,"
"But received Input(GTLabel) dim[0](%s) != "
"Input(GTBox) dim[0](%s)",
dim_gtlabel[0], dim_gtbox[0]));
PADDLE_ENFORCE_EQ(
dim_gtlabel[1], dim_gtbox[1],
platform::errors::InvalidArgument(
"Input(GTBox) and Input(GTLabel) dim[1] should be same,"
"But received Input(GTBox) dim[1](%s) != Input(GTLabel) "
"dim[1](%s)",
dim_gtbox[1], dim_gtlabel[1]));
PADDLE_ENFORCE_GT(anchors.size(), 0,
platform::errors::InvalidArgument(
"Attr(anchors) length should be greater then 0."
"But received anchors length(%s)",
anchors.size()));
PADDLE_ENFORCE_EQ(anchors.size() % 2, 0,
platform::errors::InvalidArgument(
"Attr(anchors) length should be even integer."
"But received anchors length(%s)",
anchors.size()));
for (size_t i = 0; i < anchor_mask.size(); i++) {
PADDLE_ENFORCE_LT(
anchor_mask[i], anchor_num,
platform::errors::InvalidArgument(
"Attr(anchor_mask) should not crossover Attr(anchors)."
"But received anchor_mask[i](%s) > anchor_num(%s)",
anchor_mask[i], anchor_num));
}
PADDLE_ENFORCE_GT(class_num, 0,
platform::errors::InvalidArgument(
"Attr(class_num) should be an integer greater then 0."
"But received class_num(%s) < 0",
class_num));
if (ctx->HasInput("GTScore")) {
auto dim_gtscore = ctx->GetInputDim("GTScore");
PADDLE_ENFORCE_EQ(dim_gtscore.size(), 2,
platform::errors::InvalidArgument(
"Input(GTScore) should be a 2-D tensor"
"But received GTScore dimension(%s)",
dim_gtbox.size()));
PADDLE_ENFORCE_EQ(
dim_gtscore[0], dim_gtbox[0],
platform::errors::InvalidArgument(
"Input(GTBox) and Input(GTScore) dim[0] should be same"
"But received GTBox dim[0](%s) != GTScore dim[0](%s)",
dim_gtbox[0], dim_gtscore[0]));
PADDLE_ENFORCE_EQ(
dim_gtscore[1], dim_gtbox[1],
platform::errors::InvalidArgument(
"Input(GTBox) and Input(GTScore) dim[1] should be same"
"But received GTBox dim[1](%s) != GTScore dim[1](%s)",
dim_gtscore[1], dim_gtbox[1]));
}
std::vector<int64_t> dim_out({dim_x[0]});
ctx->SetOutputDim("Loss", phi::make_ddim(dim_out));
std::vector<int64_t> dim_obj_mask({dim_x[0], mask_num, dim_x[2], dim_x[3]});
ctx->SetOutputDim("ObjectnessMask", phi::make_ddim(dim_obj_mask));
std::vector<int64_t> dim_gt_match_mask({dim_gtbox[0], dim_gtbox[1]});
ctx->SetOutputDim("GTMatchMask", phi::make_ddim(dim_gt_match_mask));
}
protected: protected:
framework::OpKernelType GetExpectedKernelType( framework::OpKernelType GetExpectedKernelType(
...@@ -347,11 +228,10 @@ class Yolov3LossGradMaker : public framework::SingleGradOpMaker<T> { ...@@ -347,11 +228,10 @@ class Yolov3LossGradMaker : public framework::SingleGradOpMaker<T> {
} // namespace paddle } // namespace paddle
namespace ops = paddle::operators; namespace ops = paddle::operators;
DECLARE_INFER_SHAPE_FUNCTOR(yolov3_loss, Yolov3LossInferShapeFunctor,
PD_INFER_META(phi::Yolov3LossInferMeta));
REGISTER_OPERATOR(yolov3_loss, ops::Yolov3LossOp, ops::Yolov3LossOpMaker, REGISTER_OPERATOR(yolov3_loss, ops::Yolov3LossOp, ops::Yolov3LossOpMaker,
ops::Yolov3LossGradMaker<paddle::framework::OpDesc>, ops::Yolov3LossGradMaker<paddle::framework::OpDesc>,
ops::Yolov3LossGradMaker<paddle::imperative::OpBase>); ops::Yolov3LossGradMaker<paddle::imperative::OpBase>,
Yolov3LossInferShapeFunctor);
REGISTER_OPERATOR(yolov3_loss_grad, ops::Yolov3LossOpGrad); REGISTER_OPERATOR(yolov3_loss_grad, ops::Yolov3LossOpGrad);
REGISTER_OP_CPU_KERNEL(yolov3_loss, ops::Yolov3LossKernel<float>,
ops::Yolov3LossKernel<double>);
REGISTER_OP_CPU_KERNEL(yolov3_loss_grad, ops::Yolov3LossGradKernel<float>,
ops::Yolov3LossGradKernel<double>);
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
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. */
#pragma once
#include <algorithm>
#include <vector>
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/phi/kernels/funcs/math_function.h"
namespace paddle {
namespace operators {
using Tensor = framework::Tensor;
template <typename T, size_t D, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenTensor = framework::EigenTensor<T, D, MajorType, IndexType>;
template <typename T, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
template <typename T>
static inline bool LessEqualZero(T x) {
return x < 1e-6;
}
template <typename T>
static T SigmoidCrossEntropy(T x, T label) {
return (x > 0 ? x : 0.0) - x * label + std::log(1.0 + std::exp(-std::abs(x)));
}
template <typename T>
static T L1Loss(T x, T y) {
return std::abs(y - x);
}
template <typename T>
static T SigmoidCrossEntropyGrad(T x, T label) {
return 1.0 / (1.0 + std::exp(-x)) - label;
}
template <typename T>
static T L1LossGrad(T x, T y) {
return x > y ? 1.0 : -1.0;
}
static int GetMaskIndex(std::vector<int> mask, int val) {
for (size_t i = 0; i < mask.size(); i++) {
if (mask[i] == val) {
return i;
}
}
return -1;
}
template <typename T>
struct Box {
T x, y, w, h;
};
template <typename T>
static inline T sigmoid(T x) {
return 1.0 / (1.0 + std::exp(-x));
}
template <typename T>
static inline Box<T> GetYoloBox(const T* x, std::vector<int> anchors, int i,
int j, int an_idx, int grid_size,
int input_size, int index, int stride,
float scale, float bias) {
Box<T> b;
b.x = (i + sigmoid<T>(x[index]) * scale + bias) / grid_size;
b.y = (j + sigmoid<T>(x[index + stride]) * scale + bias) / grid_size;
b.w = std::exp(x[index + 2 * stride]) * anchors[2 * an_idx] / input_size;
b.h = std::exp(x[index + 3 * stride]) * anchors[2 * an_idx + 1] / input_size;
return b;
}
template <typename T>
static inline Box<T> GetGtBox(const T* gt, int batch, int max_boxes, int idx) {
Box<T> b;
b.x = gt[(batch * max_boxes + idx) * 4];
b.y = gt[(batch * max_boxes + idx) * 4 + 1];
b.w = gt[(batch * max_boxes + idx) * 4 + 2];
b.h = gt[(batch * max_boxes + idx) * 4 + 3];
return b;
}
template <typename T>
static inline T BoxOverlap(T c1, T w1, T c2, T w2) {
T l1 = c1 - w1 / 2.0;
T l2 = c2 - w2 / 2.0;
T left = l1 > l2 ? l1 : l2;
T r1 = c1 + w1 / 2.0;
T r2 = c2 + w2 / 2.0;
T right = r1 < r2 ? r1 : r2;
return right - left;
}
template <typename T>
static inline T CalcBoxIoU(Box<T> b1, Box<T> b2) {
T w = BoxOverlap(b1.x, b1.w, b2.x, b2.w);
T h = BoxOverlap(b1.y, b1.h, b2.y, b2.h);
T inter_area = (w < 0 || h < 0) ? 0.0 : w * h;
T union_area = b1.w * b1.h + b2.w * b2.h - inter_area;
return inter_area / union_area;
}
static inline int GetEntryIndex(int batch, int an_idx, int hw_idx, int an_num,
int an_stride, int stride, int entry) {
return (batch * an_num + an_idx) * an_stride + entry * stride + hw_idx;
}
template <typename T>
static void CalcBoxLocationLoss(T* loss, const T* input, Box<T> gt,
std::vector<int> anchors, int an_idx,
int box_idx, int gi, int gj, int grid_size,
int input_size, int stride, T score) {
T tx = gt.x * grid_size - gi;
T ty = gt.y * grid_size - gj;
T tw = std::log(gt.w * input_size / anchors[2 * an_idx]);
T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]);
T scale = (2.0 - gt.w * gt.h) * score;
loss[0] += SigmoidCrossEntropy<T>(input[box_idx], tx) * scale;
loss[0] += SigmoidCrossEntropy<T>(input[box_idx + stride], ty) * scale;
loss[0] += L1Loss<T>(input[box_idx + 2 * stride], tw) * scale;
loss[0] += L1Loss<T>(input[box_idx + 3 * stride], th) * scale;
}
template <typename T>
static void CalcBoxLocationLossGrad(T* input_grad, const T loss, const T* input,
Box<T> gt, std::vector<int> anchors,
int an_idx, int box_idx, int gi, int gj,
int grid_size, int input_size, int stride,
T score) {
T tx = gt.x * grid_size - gi;
T ty = gt.y * grid_size - gj;
T tw = std::log(gt.w * input_size / anchors[2 * an_idx]);
T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]);
T scale = (2.0 - gt.w * gt.h) * score;
input_grad[box_idx] =
SigmoidCrossEntropyGrad<T>(input[box_idx], tx) * scale * loss;
input_grad[box_idx + stride] =
SigmoidCrossEntropyGrad<T>(input[box_idx + stride], ty) * scale * loss;
input_grad[box_idx + 2 * stride] =
L1LossGrad<T>(input[box_idx + 2 * stride], tw) * scale * loss;
input_grad[box_idx + 3 * stride] =
L1LossGrad<T>(input[box_idx + 3 * stride], th) * scale * loss;
}
template <typename T>
static inline void CalcLabelLoss(T* loss, const T* input, const int index,
const int label, const int class_num,
const int stride, const T pos, const T neg,
T score) {
for (int i = 0; i < class_num; i++) {
T pred = input[index + i * stride];
loss[0] += SigmoidCrossEntropy<T>(pred, (i == label) ? pos : neg) * score;
}
}
template <typename T>
static inline void CalcLabelLossGrad(T* input_grad, const T loss,
const T* input, const int index,
const int label, const int class_num,
const int stride, const T pos, const T neg,
T score) {
for (int i = 0; i < class_num; i++) {
T pred = input[index + i * stride];
input_grad[index + i * stride] =
SigmoidCrossEntropyGrad<T>(pred, (i == label) ? pos : neg) * score *
loss;
}
}
template <typename T>
static inline void CalcObjnessLoss(T* loss, const T* input, const T* objness,
const int n, const int an_num, const int h,
const int w, const int stride,
const int an_stride) {
for (int i = 0; i < n; i++) {
for (int j = 0; j < an_num; j++) {
for (int k = 0; k < h; k++) {
for (int l = 0; l < w; l++) {
T obj = objness[k * w + l];
if (obj > 1e-5) {
// positive sample: obj = mixup score
loss[i] += SigmoidCrossEntropy<T>(input[k * w + l], 1.0) * obj;
} else if (obj > -0.5) {
// negetive sample: obj = 0
loss[i] += SigmoidCrossEntropy<T>(input[k * w + l], 0.0);
}
}
}
objness += stride;
input += an_stride;
}
}
}
template <typename T>
static inline void CalcObjnessLossGrad(T* input_grad, const T* loss,
const T* input, const T* objness,
const int n, const int an_num,
const int h, const int w,
const int stride, const int an_stride) {
for (int i = 0; i < n; i++) {
for (int j = 0; j < an_num; j++) {
for (int k = 0; k < h; k++) {
for (int l = 0; l < w; l++) {
T obj = objness[k * w + l];
if (obj > 1e-5) {
input_grad[k * w + l] =
SigmoidCrossEntropyGrad<T>(input[k * w + l], 1.0) * obj *
loss[i];
} else if (obj > -0.5) {
input_grad[k * w + l] =
SigmoidCrossEntropyGrad<T>(input[k * w + l], 0.0) * loss[i];
}
}
}
objness += stride;
input += an_stride;
input_grad += an_stride;
}
}
}
template <typename T>
static void inline GtValid(bool* valid, const T* gtbox, const int n,
const int b) {
for (int i = 0; i < n; i++) {
for (int j = 0; j < b; j++) {
if (LessEqualZero(gtbox[j * 4 + 2]) || LessEqualZero(gtbox[j * 4 + 3])) {
valid[j] = false;
} else {
valid[j] = true;
}
}
valid += b;
gtbox += b * 4;
}
}
template <typename T>
class Yolov3LossKernel : public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& ctx) const override {
auto* input = ctx.Input<Tensor>("X");
auto* gt_box = ctx.Input<Tensor>("GTBox");
auto* gt_label = ctx.Input<Tensor>("GTLabel");
auto* gt_score = ctx.Input<Tensor>("GTScore");
auto* loss = ctx.Output<Tensor>("Loss");
auto* objness_mask = ctx.Output<Tensor>("ObjectnessMask");
auto* gt_match_mask = ctx.Output<Tensor>("GTMatchMask");
auto anchors = ctx.Attr<std::vector<int>>("anchors");
auto anchor_mask = ctx.Attr<std::vector<int>>("anchor_mask");
int class_num = ctx.Attr<int>("class_num");
float ignore_thresh = ctx.Attr<float>("ignore_thresh");
int downsample_ratio = ctx.Attr<int>("downsample_ratio");
bool use_label_smooth = ctx.Attr<bool>("use_label_smooth");
float scale = ctx.Attr<float>("scale_x_y");
float bias = -0.5 * (scale - 1.);
const int n = input->dims()[0];
const int h = input->dims()[2];
const int w = input->dims()[3];
const int an_num = anchors.size() / 2;
const int mask_num = anchor_mask.size();
const int b = gt_box->dims()[1];
int input_size = downsample_ratio * h;
const int stride = h * w;
const int an_stride = (class_num + 5) * stride;
T label_pos = 1.0;
T label_neg = 0.0;
if (use_label_smooth) {
T smooth_weight = std::min(1.0 / static_cast<T>(class_num), 1.0 / 40);
label_pos = 1.0 - smooth_weight;
label_neg = smooth_weight;
}
const T* input_data = input->data<T>();
const T* gt_box_data = gt_box->data<T>();
const int* gt_label_data = gt_label->data<int>();
T* loss_data = loss->mutable_data<T>({n}, ctx.GetPlace());
memset(loss_data, 0, loss->numel() * sizeof(T));
T* obj_mask_data =
objness_mask->mutable_data<T>({n, mask_num, h, w}, ctx.GetPlace());
memset(obj_mask_data, 0, objness_mask->numel() * sizeof(T));
int* gt_match_mask_data =
gt_match_mask->mutable_data<int>({n, b}, ctx.GetPlace());
const T* gt_score_data;
Tensor gtscore;
if (!gt_score) {
gtscore.mutable_data<T>({n, b}, ctx.GetPlace());
phi::funcs::SetConstant<platform::CPUDeviceContext, T>()(
ctx.template device_context<platform::CPUDeviceContext>(), &gtscore,
static_cast<T>(1.0));
gt_score = &gtscore;
gt_score_data = gtscore.data<T>();
} else {
gt_score_data = gt_score->data<T>();
}
// calc valid gt box mask, avoid calc duplicately in following code
Tensor gt_valid_mask;
bool* gt_valid_mask_data =
gt_valid_mask.mutable_data<bool>({n, b}, ctx.GetPlace());
GtValid<T>(gt_valid_mask_data, gt_box_data, n, b);
for (int i = 0; i < n; i++) {
for (int j = 0; j < mask_num; j++) {
for (int k = 0; k < h; k++) {
for (int l = 0; l < w; l++) {
// each predict box find a best match gt box, if overlap is bigger
// then ignore_thresh, ignore the objectness loss.
int box_idx =
GetEntryIndex(i, j, k * w + l, mask_num, an_stride, stride, 0);
Box<T> pred =
GetYoloBox(input_data, anchors, l, k, anchor_mask[j], h,
input_size, box_idx, stride, scale, bias);
T best_iou = 0;
for (int t = 0; t < b; t++) {
if (!gt_valid_mask_data[i * b + t]) {
continue;
}
Box<T> gt = GetGtBox(gt_box_data, i, b, t);
T iou = CalcBoxIoU(pred, gt);
if (iou > best_iou) {
best_iou = iou;
}
}
// If best IoU is bigger then ignore_thresh,
// ignore the objectness loss.
if (best_iou > ignore_thresh) {
int obj_idx = (i * mask_num + j) * stride + k * w + l;
obj_mask_data[obj_idx] = static_cast<T>(-1);
}
// all losses should be calculated if best IoU
// is bigger then truth thresh, but currently,
// truth thresh is an unreachable value as 1.0.
}
}
}
for (int t = 0; t < b; t++) {
if (!gt_valid_mask_data[i * b + t]) {
gt_match_mask_data[i * b + t] = -1;
continue;
}
Box<T> gt = GetGtBox(gt_box_data, i, b, t);
int gi = static_cast<int>(gt.x * w);
int gj = static_cast<int>(gt.y * h);
Box<T> gt_shift = gt;
gt_shift.x = 0.0;
gt_shift.y = 0.0;
T best_iou = 0.0;
int best_n = 0;
// each gt box find a best match anchor box as positive sample,
// for positive sample, all losses should be calculated, and for
// other samples, only objectness loss is required.
for (int an_idx = 0; an_idx < an_num; an_idx++) {
Box<T> an_box;
an_box.x = 0.0;
an_box.y = 0.0;
an_box.w = anchors[2 * an_idx] / static_cast<T>(input_size);
an_box.h = anchors[2 * an_idx + 1] / static_cast<T>(input_size);
float iou = CalcBoxIoU<T>(an_box, gt_shift);
if (iou > best_iou) {
best_iou = iou;
best_n = an_idx;
}
}
int mask_idx = GetMaskIndex(anchor_mask, best_n);
gt_match_mask_data[i * b + t] = mask_idx;
if (mask_idx >= 0) {
T score = gt_score_data[i * b + t];
int box_idx = GetEntryIndex(i, mask_idx, gj * w + gi, mask_num,
an_stride, stride, 0);
CalcBoxLocationLoss<T>(loss_data + i, input_data, gt, anchors, best_n,
box_idx, gi, gj, h, input_size, stride, score);
int obj_idx = (i * mask_num + mask_idx) * stride + gj * w + gi;
obj_mask_data[obj_idx] = score;
int label = gt_label_data[i * b + t];
int label_idx = GetEntryIndex(i, mask_idx, gj * w + gi, mask_num,
an_stride, stride, 5);
CalcLabelLoss<T>(loss_data + i, input_data, label_idx, label,
class_num, stride, label_pos, label_neg, score);
}
}
}
CalcObjnessLoss<T>(loss_data, input_data + 4 * stride, obj_mask_data, n,
mask_num, h, w, stride, an_stride);
}
};
template <typename T>
class Yolov3LossGradKernel : public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& ctx) const override {
auto* input = ctx.Input<Tensor>("X");
auto* gt_box = ctx.Input<Tensor>("GTBox");
auto* gt_label = ctx.Input<Tensor>("GTLabel");
auto* gt_score = ctx.Input<Tensor>("GTScore");
auto* input_grad = ctx.Output<Tensor>(framework::GradVarName("X"));
auto* loss_grad = ctx.Input<Tensor>(framework::GradVarName("Loss"));
auto* objness_mask = ctx.Input<Tensor>("ObjectnessMask");
auto* gt_match_mask = ctx.Input<Tensor>("GTMatchMask");
auto anchors = ctx.Attr<std::vector<int>>("anchors");
auto anchor_mask = ctx.Attr<std::vector<int>>("anchor_mask");
int class_num = ctx.Attr<int>("class_num");
int downsample_ratio = ctx.Attr<int>("downsample_ratio");
bool use_label_smooth = ctx.Attr<bool>("use_label_smooth");
const int n = input_grad->dims()[0];
const int c = input_grad->dims()[1];
const int h = input_grad->dims()[2];
const int w = input_grad->dims()[3];
const int mask_num = anchor_mask.size();
const int b = gt_match_mask->dims()[1];
int input_size = downsample_ratio * h;
const int stride = h * w;
const int an_stride = (class_num + 5) * stride;
T label_pos = 1.0;
T label_neg = 0.0;
if (use_label_smooth) {
T smooth_weight = std::min(1.0 / static_cast<T>(class_num), 1.0 / 40);
label_pos = 1.0 - smooth_weight;
label_neg = smooth_weight;
}
const T* input_data = input->data<T>();
const T* gt_box_data = gt_box->data<T>();
const int* gt_label_data = gt_label->data<int>();
const T* loss_grad_data = loss_grad->data<T>();
const T* obj_mask_data = objness_mask->data<T>();
const int* gt_match_mask_data = gt_match_mask->data<int>();
T* input_grad_data =
input_grad->mutable_data<T>({n, c, h, w}, ctx.GetPlace());
memset(input_grad_data, 0, input_grad->numel() * sizeof(T));
const T* gt_score_data;
Tensor gtscore;
if (!gt_score) {
gtscore.mutable_data<T>({n, b}, ctx.GetPlace());
phi::funcs::SetConstant<platform::CPUDeviceContext, T>()(
ctx.template device_context<platform::CPUDeviceContext>(), &gtscore,
static_cast<T>(1.0));
gt_score = &gtscore;
gt_score_data = gtscore.data<T>();
} else {
gt_score_data = gt_score->data<T>();
}
for (int i = 0; i < n; i++) {
for (int t = 0; t < b; t++) {
int mask_idx = gt_match_mask_data[i * b + t];
if (mask_idx >= 0) {
T score = gt_score_data[i * b + t];
Box<T> gt = GetGtBox(gt_box_data, i, b, t);
int gi = static_cast<int>(gt.x * w);
int gj = static_cast<int>(gt.y * h);
int box_idx = GetEntryIndex(i, mask_idx, gj * w + gi, mask_num,
an_stride, stride, 0);
CalcBoxLocationLossGrad<T>(input_grad_data, loss_grad_data[i],
input_data, gt, anchors,
anchor_mask[mask_idx], box_idx, gi, gj, h,
input_size, stride, score);
int label = gt_label_data[i * b + t];
int label_idx = GetEntryIndex(i, mask_idx, gj * w + gi, mask_num,
an_stride, stride, 5);
CalcLabelLossGrad<T>(input_grad_data, loss_grad_data[i], input_data,
label_idx, label, class_num, stride, label_pos,
label_neg, score);
}
}
}
CalcObjnessLossGrad<T>(input_grad_data + 4 * stride, loss_grad_data,
input_data + 4 * stride, obj_mask_data, n, mask_num,
h, w, stride, an_stride);
}
};
} // namespace operators
} // namespace paddle
...@@ -1229,6 +1229,153 @@ void WhereInferMeta(const MetaTensor& condition, ...@@ -1229,6 +1229,153 @@ void WhereInferMeta(const MetaTensor& condition,
out->share_meta(x); out->share_meta(x);
} }
void Yolov3LossInferMeta(const MetaTensor& x,
const MetaTensor& gt_box,
const MetaTensor& gt_label,
const paddle::optional<const MetaTensor&> gt_score,
const std::vector<int>& anchors,
const std::vector<int>& anchor_mask,
int class_num,
float ignore_thresh,
int downsample_ratio,
bool use_label_smooth,
float scale_x_y,
MetaTensor* loss,
MetaTensor* objectness_mask,
MetaTensor* gt_match_mask) {
auto dim_x = x.dims();
auto dim_gtbox = gt_box.dims();
auto dim_gtlabel = gt_label.dims();
int anchor_num = anchors.size() / 2;
int mask_num = anchor_mask.size();
PADDLE_ENFORCE_EQ(dim_x.size(),
4,
phi::errors::InvalidArgument(
"Input(X) should be a 4-D tensor. But received "
"X dimension size(%s)",
dim_x.size()));
PADDLE_ENFORCE_EQ(
dim_x[2],
dim_x[3],
phi::errors::InvalidArgument("Input(X) dim[3] and dim[4] should be euqal."
"But received dim[3](%s) != dim[4](%s)",
dim_x[2],
dim_x[3]));
PADDLE_ENFORCE_EQ(
dim_x[1],
mask_num * (5 + class_num),
phi::errors::InvalidArgument(
"Input(X) dim[1] should be equal to (anchor_mask_number * (5 "
"+ class_num))."
"But received dim[1](%s) != (anchor_mask_number * "
"(5+class_num)(%s).",
dim_x[1],
mask_num * (5 + class_num)));
PADDLE_ENFORCE_EQ(
dim_gtbox.size(),
3,
phi::errors::InvalidArgument("Input(GTBox) should be a 3-D tensor, but "
"received gtbox dimension size(%s)",
dim_gtbox.size()));
PADDLE_ENFORCE_EQ(
dim_gtbox[2],
4,
phi::errors::InvalidArgument("Input(GTBox) dim[2] should be 4",
"But receive dim[2](%s) != 5. ",
dim_gtbox[2]));
PADDLE_ENFORCE_EQ(dim_gtlabel.size(),
2,
phi::errors::InvalidArgument(
"Input(GTLabel) should be a 2-D tensor,"
"But received Input(GTLabel) dimension size(%s) != 2.",
dim_gtlabel.size()));
PADDLE_ENFORCE_EQ(
dim_gtlabel[0],
dim_gtbox[0],
phi::errors::InvalidArgument(
"Input(GTBox) dim[0] and Input(GTLabel) dim[0] should be same,"
"But received Input(GTLabel) dim[0](%s) != "
"Input(GTBox) dim[0](%s)",
dim_gtlabel[0],
dim_gtbox[0]));
PADDLE_ENFORCE_EQ(
dim_gtlabel[1],
dim_gtbox[1],
phi::errors::InvalidArgument(
"Input(GTBox) and Input(GTLabel) dim[1] should be same,"
"But received Input(GTBox) dim[1](%s) != Input(GTLabel) "
"dim[1](%s)",
dim_gtbox[1],
dim_gtlabel[1]));
PADDLE_ENFORCE_GT(anchors.size(),
0,
phi::errors::InvalidArgument(
"Attr(anchors) length should be greater then 0."
"But received anchors length(%s)",
anchors.size()));
PADDLE_ENFORCE_EQ(anchors.size() % 2,
0,
phi::errors::InvalidArgument(
"Attr(anchors) length should be even integer."
"But received anchors length(%s)",
anchors.size()));
for (size_t i = 0; i < anchor_mask.size(); i++) {
PADDLE_ENFORCE_LT(
anchor_mask[i],
anchor_num,
phi::errors::InvalidArgument(
"Attr(anchor_mask) should not crossover Attr(anchors)."
"But received anchor_mask[i](%s) > anchor_num(%s)",
anchor_mask[i],
anchor_num));
}
PADDLE_ENFORCE_GT(class_num,
0,
phi::errors::InvalidArgument(
"Attr(class_num) should be an integer greater then 0."
"But received class_num(%s) < 0",
class_num));
if (gt_score.get_ptr()) {
auto dim_gtscore = gt_score->dims();
PADDLE_ENFORCE_EQ(
dim_gtscore.size(),
2,
phi::errors::InvalidArgument("Input(GTScore) should be a 2-D tensor"
"But received GTScore dimension(%s)",
dim_gtbox.size()));
PADDLE_ENFORCE_EQ(
dim_gtscore[0],
dim_gtbox[0],
phi::errors::InvalidArgument(
"Input(GTBox) and Input(GTScore) dim[0] should be same"
"But received GTBox dim[0](%s) != GTScore dim[0](%s)",
dim_gtbox[0],
dim_gtscore[0]));
PADDLE_ENFORCE_EQ(
dim_gtscore[1],
dim_gtbox[1],
phi::errors::InvalidArgument(
"Input(GTBox) and Input(GTScore) dim[1] should be same"
"But received GTBox dim[1](%s) != GTScore dim[1](%s)",
dim_gtscore[1],
dim_gtbox[1]));
}
std::vector<int64_t> dim_out({dim_x[0]});
loss->set_dims(phi::make_ddim(dim_out));
loss->set_dtype(x.dtype());
std::vector<int64_t> dim_obj_mask({dim_x[0], mask_num, dim_x[2], dim_x[3]});
objectness_mask->set_dims(phi::make_ddim(dim_obj_mask));
objectness_mask->set_dtype(x.dtype());
std::vector<int64_t> dim_gt_match_mask({dim_gtbox[0], dim_gtbox[1]});
gt_match_mask->set_dims(phi::make_ddim(dim_gt_match_mask));
gt_match_mask->set_dtype(x.dtype());
}
} // namespace phi } // namespace phi
PD_REGISTER_INFER_META_FN(batch_norm, phi::BatchNormInferMeta); PD_REGISTER_INFER_META_FN(batch_norm, phi::BatchNormInferMeta);
......
...@@ -245,4 +245,19 @@ void WhereInferMeta(const MetaTensor& condition, ...@@ -245,4 +245,19 @@ void WhereInferMeta(const MetaTensor& condition,
const MetaTensor& y, const MetaTensor& y,
MetaTensor* out); MetaTensor* out);
void Yolov3LossInferMeta(const MetaTensor& x,
const MetaTensor& gt_box,
const MetaTensor& gt_label,
const paddle::optional<const MetaTensor&> gt_score,
const std::vector<int>& anchors,
const std::vector<int>& anchor_mask,
int class_num,
float ignore_thresh,
int downsample_ratio,
bool use_label_smooth,
float scale_x_y,
MetaTensor* loss,
MetaTensor* objectness_mask,
MetaTensor* gt_match_mask);
} // namespace phi } // namespace phi
// Copyright (c) 2022 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.
#pragma once
namespace phi {
template <typename T>
struct Box {
T x, y, w, h;
};
template <typename T>
static inline T sigmoid(T x) {
return 1.0 / (1.0 + std::exp(-x));
}
template <typename T>
static inline Box<T> GetGtBox(const T* gt, int batch, int max_boxes, int idx) {
Box<T> b;
b.x = gt[(batch * max_boxes + idx) * 4];
b.y = gt[(batch * max_boxes + idx) * 4 + 1];
b.w = gt[(batch * max_boxes + idx) * 4 + 2];
b.h = gt[(batch * max_boxes + idx) * 4 + 3];
return b;
}
static inline int GetEntryIndex(int batch,
int an_idx,
int hw_idx,
int an_num,
int an_stride,
int stride,
int entry) {
return (batch * an_num + an_idx) * an_stride + entry * stride + hw_idx;
}
} // namespace phi
// Copyright (c) 2022 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 <algorithm>
#include <vector>
#include "paddle/phi/kernels/yolov3_loss_grad_kernel.h"
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/cpu/yolov3_loss_functor.h"
#include "paddle/phi/kernels/funcs/math_function.h"
namespace phi {
template <typename T>
static T SigmoidCrossEntropyGrad(T x, T label) {
return 1.0 / (1.0 + std::exp(-x)) - label;
}
template <typename T>
static T L1LossGrad(T x, T y) {
return x > y ? 1.0 : -1.0;
}
template <typename T>
static void CalcBoxLocationLossGrad(T* input_grad,
const T loss,
const T* input,
Box<T> gt,
std::vector<int> anchors,
int an_idx,
int box_idx,
int gi,
int gj,
int grid_size,
int input_size,
int stride,
T score) {
T tx = gt.x * grid_size - gi;
T ty = gt.y * grid_size - gj;
T tw = std::log(gt.w * input_size / anchors[2 * an_idx]);
T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]);
T scale = (2.0 - gt.w * gt.h) * score;
input_grad[box_idx] =
SigmoidCrossEntropyGrad<T>(input[box_idx], tx) * scale * loss;
input_grad[box_idx + stride] =
SigmoidCrossEntropyGrad<T>(input[box_idx + stride], ty) * scale * loss;
input_grad[box_idx + 2 * stride] =
L1LossGrad<T>(input[box_idx + 2 * stride], tw) * scale * loss;
input_grad[box_idx + 3 * stride] =
L1LossGrad<T>(input[box_idx + 3 * stride], th) * scale * loss;
}
template <typename T>
static inline void CalcLabelLossGrad(T* input_grad,
const T loss,
const T* input,
const int index,
const int label,
const int class_num,
const int stride,
const T pos,
const T neg,
T score) {
for (int i = 0; i < class_num; i++) {
T pred = input[index + i * stride];
input_grad[index + i * stride] =
SigmoidCrossEntropyGrad<T>(pred, (i == label) ? pos : neg) * score *
loss;
}
}
template <typename T>
static inline void CalcObjnessLossGrad(T* input_grad,
const T* loss,
const T* input,
const T* objness,
const int n,
const int an_num,
const int h,
const int w,
const int stride,
const int an_stride) {
for (int i = 0; i < n; i++) {
for (int j = 0; j < an_num; j++) {
for (int k = 0; k < h; k++) {
for (int l = 0; l < w; l++) {
T obj = objness[k * w + l];
if (obj > 1e-5) {
input_grad[k * w + l] =
SigmoidCrossEntropyGrad<T>(input[k * w + l], 1.0) * obj *
loss[i];
} else if (obj > -0.5) {
input_grad[k * w + l] =
SigmoidCrossEntropyGrad<T>(input[k * w + l], 0.0) * loss[i];
}
}
}
objness += stride;
input += an_stride;
input_grad += an_stride;
}
}
}
template <typename T, typename Context>
void Yolov3LossGradKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& gt_box,
const DenseTensor& gt_label,
paddle::optional<const DenseTensor&> gt_score,
const DenseTensor& loss_grad,
const DenseTensor& objectness_mask,
const DenseTensor& gt_match_mask,
const std::vector<int>& anchors,
const std::vector<int>& anchor_mask,
int class_num,
float ignore_thresh,
int downsample_ratio,
bool use_label_smooth,
float scale_x_y,
DenseTensor* x_grad,
DenseTensor* gt_box_grad,
DenseTensor* gt_label_grad,
DenseTensor* gt_score_grad) {
auto* input = &x;
auto input_grad = x_grad;
auto* objness_mask = &objectness_mask;
const int n = input_grad->dims()[0];
const int c = input_grad->dims()[1];
const int h = input_grad->dims()[2];
const int w = input_grad->dims()[3];
const int mask_num = anchor_mask.size();
const int b = gt_match_mask.dims()[1];
int input_size = downsample_ratio * h;
const int stride = h * w;
const int an_stride = (class_num + 5) * stride;
T label_pos = 1.0;
T label_neg = 0.0;
if (use_label_smooth) {
T smooth_weight = std::min(1.0 / static_cast<T>(class_num), 1.0 / 40);
label_pos = 1.0 - smooth_weight;
label_neg = smooth_weight;
}
const T* input_data = input->data<T>();
const T* gt_box_data = gt_box.data<T>();
const int* gt_label_data = gt_label.data<int>();
const T* loss_grad_data = loss_grad.data<T>();
const T* obj_mask_data = objness_mask->data<T>();
const int* gt_match_mask_data = gt_match_mask.data<int>();
input_grad->Resize({n, c, h, w});
T* input_grad_data = dev_ctx.template Alloc<T>(input_grad);
memset(input_grad_data, 0, input_grad->numel() * sizeof(T));
const T* gt_score_data;
DenseTensor gtscore;
if (!(gt_score.is_initialized())) {
gtscore.Resize({n, b});
dev_ctx.template Alloc<T>(&gtscore);
phi::funcs::SetConstant<Context, T>()(
dev_ctx, &gtscore, static_cast<T>(1.0));
gt_score_data = gtscore.data<T>();
} else {
gt_score_data = gt_score.get_ptr()->data<T>();
}
for (int i = 0; i < n; i++) {
for (int t = 0; t < b; t++) {
int mask_idx = gt_match_mask_data[i * b + t];
if (mask_idx >= 0) {
T score = gt_score_data[i * b + t];
Box<T> gt = GetGtBox(gt_box_data, i, b, t);
int gi = static_cast<int>(gt.x * w);
int gj = static_cast<int>(gt.y * h);
int box_idx = GetEntryIndex(
i, mask_idx, gj * w + gi, mask_num, an_stride, stride, 0);
CalcBoxLocationLossGrad<T>(input_grad_data,
loss_grad_data[i],
input_data,
gt,
anchors,
anchor_mask[mask_idx],
box_idx,
gi,
gj,
h,
input_size,
stride,
score);
int label = gt_label_data[i * b + t];
int label_idx = GetEntryIndex(
i, mask_idx, gj * w + gi, mask_num, an_stride, stride, 5);
CalcLabelLossGrad<T>(input_grad_data,
loss_grad_data[i],
input_data,
label_idx,
label,
class_num,
stride,
label_pos,
label_neg,
score);
}
}
}
CalcObjnessLossGrad<T>(input_grad_data + 4 * stride,
loss_grad_data,
input_data + 4 * stride,
obj_mask_data,
n,
mask_num,
h,
w,
stride,
an_stride);
}
} // namespace phi
PD_REGISTER_KERNEL(yolov3_loss_grad,
CPU,
ALL_LAYOUT,
phi::Yolov3LossGradKernel,
float,
double) {}
// Copyright (c) 2022 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 <algorithm>
#include <vector>
#include "paddle/phi/kernels/yolov3_loss_kernel.h"
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/cpu/yolov3_loss_functor.h"
#include "paddle/phi/kernels/funcs/math_function.h"
namespace phi {
template <typename T>
static inline bool LessEqualZero(T x) {
return x < 1e-6;
}
template <typename T>
static T SigmoidCrossEntropy(T x, T label) {
return (x > 0 ? x : 0.0) - x * label + std::log(1.0 + std::exp(-std::abs(x)));
}
template <typename T>
static T L1Loss(T x, T y) {
return std::abs(y - x);
}
static int GetMaskIndex(std::vector<int> mask, int val) {
for (size_t i = 0; i < mask.size(); i++) {
if (mask[i] == val) {
return i;
}
}
return -1;
}
template <typename T>
static inline Box<T> GetYoloBox(const T* x,
std::vector<int> anchors,
int i,
int j,
int an_idx,
int grid_size,
int input_size,
int index,
int stride,
float scale,
float bias) {
Box<T> b;
b.x = (i + sigmoid<T>(x[index]) * scale + bias) / grid_size;
b.y = (j + sigmoid<T>(x[index + stride]) * scale + bias) / grid_size;
b.w = std::exp(x[index + 2 * stride]) * anchors[2 * an_idx] / input_size;
b.h = std::exp(x[index + 3 * stride]) * anchors[2 * an_idx + 1] / input_size;
return b;
}
template <typename T>
static inline T BoxOverlap(T c1, T w1, T c2, T w2) {
T l1 = c1 - w1 / 2.0;
T l2 = c2 - w2 / 2.0;
T left = l1 > l2 ? l1 : l2;
T r1 = c1 + w1 / 2.0;
T r2 = c2 + w2 / 2.0;
T right = r1 < r2 ? r1 : r2;
return right - left;
}
template <typename T>
static inline T CalcBoxIoU(Box<T> b1, Box<T> b2) {
T w = BoxOverlap(b1.x, b1.w, b2.x, b2.w);
T h = BoxOverlap(b1.y, b1.h, b2.y, b2.h);
T inter_area = (w < 0 || h < 0) ? 0.0 : w * h;
T union_area = b1.w * b1.h + b2.w * b2.h - inter_area;
return inter_area / union_area;
}
template <typename T>
static void CalcBoxLocationLoss(T* loss,
const T* input,
Box<T> gt,
std::vector<int> anchors,
int an_idx,
int box_idx,
int gi,
int gj,
int grid_size,
int input_size,
int stride,
T score) {
T tx = gt.x * grid_size - gi;
T ty = gt.y * grid_size - gj;
T tw = std::log(gt.w * input_size / anchors[2 * an_idx]);
T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]);
T scale = (2.0 - gt.w * gt.h) * score;
loss[0] += SigmoidCrossEntropy<T>(input[box_idx], tx) * scale;
loss[0] += SigmoidCrossEntropy<T>(input[box_idx + stride], ty) * scale;
loss[0] += L1Loss<T>(input[box_idx + 2 * stride], tw) * scale;
loss[0] += L1Loss<T>(input[box_idx + 3 * stride], th) * scale;
}
template <typename T>
static inline void CalcLabelLoss(T* loss,
const T* input,
const int index,
const int label,
const int class_num,
const int stride,
const T pos,
const T neg,
T score) {
for (int i = 0; i < class_num; i++) {
T pred = input[index + i * stride];
loss[0] += SigmoidCrossEntropy<T>(pred, (i == label) ? pos : neg) * score;
}
}
template <typename T>
static inline void CalcObjnessLoss(T* loss,
const T* input,
const T* objness,
const int n,
const int an_num,
const int h,
const int w,
const int stride,
const int an_stride) {
for (int i = 0; i < n; i++) {
for (int j = 0; j < an_num; j++) {
for (int k = 0; k < h; k++) {
for (int l = 0; l < w; l++) {
T obj = objness[k * w + l];
if (obj > 1e-5) {
// positive sample: obj = mixup score
loss[i] += SigmoidCrossEntropy<T>(input[k * w + l], 1.0) * obj;
} else if (obj > -0.5) {
// negetive sample: obj = 0
loss[i] += SigmoidCrossEntropy<T>(input[k * w + l], 0.0);
}
}
}
objness += stride;
input += an_stride;
}
}
}
template <typename T>
static void inline GtValid(bool* valid,
const T* gtbox,
const int n,
const int b) {
for (int i = 0; i < n; i++) {
for (int j = 0; j < b; j++) {
if (LessEqualZero(gtbox[j * 4 + 2]) || LessEqualZero(gtbox[j * 4 + 3])) {
valid[j] = false;
} else {
valid[j] = true;
}
}
valid += b;
gtbox += b * 4;
}
}
template <typename T, typename Context>
void Yolov3LossKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& gt_box,
const DenseTensor& gt_label,
paddle::optional<const DenseTensor&> gt_score,
const std::vector<int>& anchors,
const std::vector<int>& anchor_mask,
int class_num,
float ignore_thresh,
int downsample_ratio,
bool use_label_smooth,
float scale_x_y,
DenseTensor* loss,
DenseTensor* objectness_mask,
DenseTensor* gt_match_mask) {
auto* input = &x;
auto objness_mask = objectness_mask;
float scale = scale_x_y;
float bias = -0.5 * (scale - 1.);
const int n = input->dims()[0];
const int h = input->dims()[2];
const int w = input->dims()[3];
const int an_num = anchors.size() / 2;
const int mask_num = anchor_mask.size();
const int b = gt_box.dims()[1];
int input_size = downsample_ratio * h;
const int stride = h * w;
const int an_stride = (class_num + 5) * stride;
T label_pos = 1.0;
T label_neg = 0.0;
if (use_label_smooth) {
T smooth_weight = std::min(1.0 / static_cast<T>(class_num), 1.0 / 40);
label_pos = 1.0 - smooth_weight;
label_neg = smooth_weight;
}
const T* input_data = input->data<T>();
const T* gt_box_data = gt_box.data<T>();
const int* gt_label_data = gt_label.data<int>();
loss->Resize({n});
T* loss_data = dev_ctx.template Alloc<T>(loss);
memset(loss_data, 0, loss->numel() * sizeof(T));
objness_mask->Resize({n, mask_num, h, w});
T* obj_mask_data = dev_ctx.template Alloc<T>(objness_mask);
memset(obj_mask_data, 0, objness_mask->numel() * sizeof(T));
gt_match_mask->Resize({n, b});
int* gt_match_mask_data = dev_ctx.template Alloc<int>(gt_match_mask);
const T* gt_score_data;
DenseTensor gtscore;
if (!(gt_score.is_initialized())) {
gtscore.Resize({n, b});
dev_ctx.template Alloc<T>(&gtscore);
phi::funcs::SetConstant<Context, T>()(
dev_ctx, &gtscore, static_cast<T>(1.0));
gt_score_data = gtscore.data<T>();
} else {
gt_score_data = gt_score.get_ptr()->data<T>();
}
// calc valid gt box mask, avoid calc duplicately in following code
DenseTensor gt_valid_mask;
gt_valid_mask.Resize({n, b});
bool* gt_valid_mask_data = dev_ctx.template Alloc<bool>(&gt_valid_mask);
GtValid<T>(gt_valid_mask_data, gt_box_data, n, b);
for (int i = 0; i < n; i++) {
for (int j = 0; j < mask_num; j++) {
for (int k = 0; k < h; k++) {
for (int l = 0; l < w; l++) {
// each predict box find a best match gt box, if overlap is bigger
// then ignore_thresh, ignore the objectness loss.
int box_idx =
GetEntryIndex(i, j, k * w + l, mask_num, an_stride, stride, 0);
Box<T> pred = GetYoloBox(input_data,
anchors,
l,
k,
anchor_mask[j],
h,
input_size,
box_idx,
stride,
scale,
bias);
T best_iou = 0;
for (int t = 0; t < b; t++) {
if (!gt_valid_mask_data[i * b + t]) {
continue;
}
Box<T> gt = GetGtBox(gt_box_data, i, b, t);
T iou = CalcBoxIoU(pred, gt);
if (iou > best_iou) {
best_iou = iou;
}
}
// If best IoU is bigger then ignore_thresh,
// ignore the objectness loss.
if (best_iou > ignore_thresh) {
int obj_idx = (i * mask_num + j) * stride + k * w + l;
obj_mask_data[obj_idx] = static_cast<T>(-1);
}
// all losses should be calculated if best IoU
// is bigger then truth thresh, but currently,
// truth thresh is an unreachable value as 1.0.
}
}
}
for (int t = 0; t < b; t++) {
if (!gt_valid_mask_data[i * b + t]) {
gt_match_mask_data[i * b + t] = -1;
continue;
}
Box<T> gt = GetGtBox(gt_box_data, i, b, t);
int gi = static_cast<int>(gt.x * w);
int gj = static_cast<int>(gt.y * h);
Box<T> gt_shift = gt;
gt_shift.x = 0.0;
gt_shift.y = 0.0;
T best_iou = 0.0;
int best_n = 0;
// each gt box find a best match anchor box as positive sample,
// for positive sample, all losses should be calculated, and for
// other samples, only objectness loss is required.
for (int an_idx = 0; an_idx < an_num; an_idx++) {
Box<T> an_box;
an_box.x = 0.0;
an_box.y = 0.0;
an_box.w = anchors[2 * an_idx] / static_cast<T>(input_size);
an_box.h = anchors[2 * an_idx + 1] / static_cast<T>(input_size);
float iou = CalcBoxIoU<T>(an_box, gt_shift);
if (iou > best_iou) {
best_iou = iou;
best_n = an_idx;
}
}
int mask_idx = GetMaskIndex(anchor_mask, best_n);
gt_match_mask_data[i * b + t] = mask_idx;
if (mask_idx >= 0) {
T score = gt_score_data[i * b + t];
int box_idx = GetEntryIndex(
i, mask_idx, gj * w + gi, mask_num, an_stride, stride, 0);
CalcBoxLocationLoss<T>(loss_data + i,
input_data,
gt,
anchors,
best_n,
box_idx,
gi,
gj,
h,
input_size,
stride,
score);
int obj_idx = (i * mask_num + mask_idx) * stride + gj * w + gi;
obj_mask_data[obj_idx] = score;
int label = gt_label_data[i * b + t];
int label_idx = GetEntryIndex(
i, mask_idx, gj * w + gi, mask_num, an_stride, stride, 5);
CalcLabelLoss<T>(loss_data + i,
input_data,
label_idx,
label,
class_num,
stride,
label_pos,
label_neg,
score);
}
}
}
CalcObjnessLoss<T>(loss_data,
input_data + 4 * stride,
obj_mask_data,
n,
mask_num,
h,
w,
stride,
an_stride);
}
} // namespace phi
PD_REGISTER_KERNEL(
yolov3_loss, CPU, ALL_LAYOUT, phi::Yolov3LossKernel, float, double) {}
// Copyright (c) 2022 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.
#pragma once
#include "paddle/phi/core/dense_tensor.h"
namespace phi {
template <typename T, typename Context>
void Yolov3LossGradKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& gt_box,
const DenseTensor& gt_label,
paddle::optional<const DenseTensor&> gt_score,
const DenseTensor& loss_grad,
const DenseTensor& objectness_mask,
const DenseTensor& gt_match_mask,
const std::vector<int>& anchors,
const std::vector<int>& anchor_mask,
int class_num,
float ignore_thresh,
int downsample_ratio,
bool use_label_smooth,
float scale_x_Y,
DenseTensor* x_grad,
DenseTensor* gt_box_grad,
DenseTensor* gt_label_grad,
DenseTensor* gt_score_grad);
} // namespace phi
// Copyright (c) 2022 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.
#pragma once
#include "paddle/phi/core/dense_tensor.h"
namespace phi {
template <typename T, typename Context>
void Yolov3LossKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& gt_box,
const DenseTensor& gt_label,
paddle::optional<const DenseTensor&> gt_score,
const std::vector<int>& anchors,
const std::vector<int>& anchor_mask,
int class_num,
float ignore_thresh,
int downsample_ratio,
bool use_label_smooth,
float scale_x_Y,
DenseTensor* loss,
DenseTensor* objectness_mask,
DenseTensor* gt_match_mask);
} // namespace phi
// Copyright (c) 2022 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/core/compat/op_utils.h"
namespace phi {
KernelSignature Yolov3LossOpArgumentMapping(const ArgumentMappingContext& ctx) {
return KernelSignature("yolov3_loss",
{"X", "GTBox", "GTLabel", "GTScore"},
{"anchors",
"anchor_mask",
"class_num",
"ignore_thresh",
"downsample_ratio",
"use_label_smooth",
"scale_x_y"},
{"Loss", "ObjectnessMask", "GTMatchMask"});
}
KernelSignature Yolov3LossGradOpArgumentMapping(
const ArgumentMappingContext& ctx) {
return KernelSignature("yolov3_loss_grad",
{"X",
"GTBox",
"GTLabel",
"GTScore",
GradVarName("Loss"),
"ObjectnessMask",
"GTMatchMask"},
{"anchors",
"anchor_mask",
"class_num",
"ignore_thresh",
"downsample_ratio",
"use_label_smooth",
"scale_x_y"},
{GradVarName("X"),
GradVarName("GTBox"),
GradVarName("GTLabel"),
GradVarName("GTScore")});
}
} // namespace phi
PD_REGISTER_ARG_MAPPING_FN(yolov3_loss, phi::Yolov3LossOpArgumentMapping);
PD_REGISTER_ARG_MAPPING_FN(yolov3_loss_grad,
phi::Yolov3LossGradOpArgumentMapping);
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册