Skip to content

P
Paddle

机器未来 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 1a685144 编写于 作者: S sweetsky0901
浏览文件
操作

for xxYY to xx_yy

上级 dc7ddcb0
显示空白变更内容
内联 并排
Showing with 56 addition and 57 deletion
paddle/operators/detection_output_op.cc
浏览文件 @ 1a685144
......@@ -18,8 +18,7 @@ namespace operators {
class DetectionOutputOpMaker : public framework::OpProtoAndCheckerMaker {
public:
DetectionOutputOpMaker(framework::OpProto* proto,
framework::OpAttrChecker* op_checker)
DetectionOutputOpMaker(OpProto* proto, OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddInput("Loc",
"(Tensor) The input tensor of detection_output operator."
......@@ -86,5 +85,5 @@ REGISTER_OP_WITHOUT_GRADIENT(detection_output, ops::DetectionOutputOp,
ops::DetectionOutputOpMaker);
REGISTER_OP_CPU_KERNEL(
detection_output,
ops::Detection_output_Kernel<paddle::platform::CPUDeviceContext, float>,
ops::Detection_output_Kernel<paddle::platform::CPUDeviceContext, double>);
ops::DetectionOutputKernel<paddle::platform::CPUDeviceContext, float>,
ops::DetectionOutputKernel<paddle::platform::CPUDeviceContext, double>);
paddle/operators/detection_output_op.cu.cc
浏览文件 @ 1a685144
......@@ -17,5 +17,5 @@ limitations under the License. */
namespace ops = paddle::operators;
REGISTER_OP_CUDA_KERNEL(
detection_output,
ops::Detection_output_Kernel<paddle::platform::CUDADeviceContext, float>,
ops::Detection_output_Kernel<paddle::platform::CUDADeviceContext, double>);
ops::DetectionOutputKernel<paddle::platform::CUDADeviceContext, float>,
ops::DetectionOutputKernel<paddle::platform::CUDADeviceContext, double>);
paddle/operators/detection_output_op.h
浏览文件 @ 1a685144
......@@ -48,7 +48,7 @@ inline void transpose_fun(const framework::ExecutionContext& context,
}
}
template <typename DeviceContext, typename T>
class Detection_output_Kernel : public framework::OpKernel<T> {
class DetectionOutputKernel : public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& context) const override {
const framework::Tensor* in_loc = context.Input<framework::Tensor>("Loc");
......@@ -119,22 +119,22 @@ class Detection_output_Kernel : public framework::OpKernel<T> {
size_t prior_offset = i * 8;
size_t loc_pred_offset = n * num_priors * 4 + i * 4;
std::vector<math::BBox<T>> prior_bbox_vec;
math::getBBoxFromPriorData<T>(priorbox_data + prior_offset, 1,
math::get_bbox_from_priorData<T>(priorbox_data + prior_offset, 1,
prior_bbox_vec);
std::vector<std::vector<T>> prior_bbox_var;
math::getBBoxVarFromPriorData<T>(priorbox_data + prior_offset, 1,
math::get_bbox_var_from_prior_data<T>(priorbox_data + prior_offset, 1,
prior_bbox_var);
std::vector<T> loc_pred_data;
for (size_t j = 0; j < 4; ++j)
loc_pred_data.push_back(*(loc_data + loc_pred_offset + j));
math::BBox<T> bbox = math::decodeBBoxWithVar<T>(
math::BBox<T> bbox = math::decode_bbox_with_var<T>(
prior_bbox_vec[0], prior_bbox_var[0], loc_pred_data);
decoded_bboxes.push_back(bbox);
}
all_decoded_bboxes.push_back(decoded_bboxes);
}
std::vector<std::map<size_t, std::vector<size_t>>> all_indices;
int num_kept = math::getDetectionIndices<T>(
int num_kept = math::get_detection_indices<T>(
conf_data, num_priors, num_classes, background_label_id, batch_size,
confidence_threshold, nms_top_k, nms_threshold, top_k,
all_decoded_bboxes, &all_indices);
......@@ -154,11 +154,11 @@ class Detection_output_Kernel : public framework::OpKernel<T> {
out_cpu.mutable_data<T>(out->dims(), platform::CPUPlace());
out_data = out_cpu.data<T>();
}
math::getDetectionOutput<T>(conf_data, num_kept, num_priors, num_classes,
math::get_detection_output<T>(conf_data, num_kept, num_priors, num_classes,
batch_size, all_indices, all_decoded_bboxes,
out_data);
if (platform::is_gpu_place(context.GetPlace())) {
framework::CopyFrom(out_cpu, platform::GPUPlace(),
framework::CopyFrom(out_cpu, platform::CUDAPlace(),
context.device_context(), out);
}
}
......
paddle/operators/math/detection_util.h
浏览文件 @ 1a685144
......@@ -51,31 +51,31 @@ struct BBox {
// KNCHW ==> NHWC
// template <typename T>
template <typename T>
void getBBoxFromPriorData(const T* prior_data, const size_t num_bboxes,
void get_bbox_from_priorData(const T* prior_data, const size_t num_bboxes,
std::vector<BBox<T>>& bbox_vec);
template <typename T>
void getBBoxVarFromPriorData(const T* prior_data, const size_t num,
void get_bbox_var_from_prior_data(const T* prior_data, const size_t num,
std::vector<std::vector<T>>& var_vec);
template <typename T>
BBox<T> decodeBBoxWithVar(BBox<T>& prior_bbox,
BBox<T> decode_bbox_with_var(BBox<T>& prior_bbox,
const std::vector<T>& prior_bbox_var,
const std::vector<T>& loc_pred_data);
template <typename T1, typename T2>
bool sortScorePairDescend(const std::pair<T1, T2>& pair1,
bool sort_score_pair_descend(const std::pair<T1, T2>& pair1,
const std::pair<T1, T2>& pair2);
template <typename T>
bool sortScorePairDescend(const std::pair<T, BBox<T>>& pair1,
bool sort_score_pair_descend(const std::pair<T, BBox<T>>& pair1,
const std::pair<T, BBox<T>>& pair2);
template <typename T>
T jaccardOverlap(const BBox<T>& bbox1, const BBox<T>& bbox2);
T jaccard_overlap(const BBox<T>& bbox1, const BBox<T>& bbox2);
template <typename T>
void applyNMSFast(const std::vector<BBox<T>>& bboxes, const T* conf_score_data,
size_t class_idx, size_t top_k, T conf_threshold,
T nms_threshold, size_t num_priors, size_t num_classes,
std::vector<size_t>* indices);
void apply_nms_fast(const std::vector<BBox<T>>& bboxes,
const T* conf_score_data, size_t class_idx, size_t top_k,
T conf_threshold, T nms_threshold, size_t num_priors,
size_t num_classes, std::vector<size_t>* indices);
template <typename T>
int getDetectionIndices(
int get_detection_indices(
const T* conf_data, const size_t num_priors, const size_t num_classes,
const size_t background_label_id, const size_t batch_size,
const T conf_threshold, const size_t nms_top_k, const T nms_threshold,
......@@ -85,13 +85,13 @@ int getDetectionIndices(
template <typename T>
BBox<T> clipBBox(const BBox<T>& bbox);
template <typename T>
void getDetectionOutput(
void get_detection_output(
const T* conf_data, const size_t num_kept, const size_t num_priors,
const size_t num_classes, const size_t batch_size,
const std::vector<std::map<size_t, std::vector<size_t>>>& all_indices,
const std::vector<std::vector<BBox<T>>>& all_decoded_bboxes, T* out_data);
template <typename T>
void getBBoxFromPriorData(const T* prior_data, const size_t num_bboxes,
void get_bbox_from_priorData(const T* prior_data, const size_t num_bboxes,
std::vector<BBox<T>>& bbox_vec) {
size_t out_offset = bbox_vec.size();
bbox_vec.resize(bbox_vec.size() + num_bboxes);
......@@ -105,7 +105,7 @@ void getBBoxFromPriorData(const T* prior_data, const size_t num_bboxes,
}
}
template <typename T>
void getBBoxVarFromPriorData(const T* prior_data, const size_t num,
void get_bbox_var_from_prior_data(const T* prior_data, const size_t num,
std::vector<std::vector<T>>& var_vec) {
size_t out_offset = var_vec.size();
var_vec.resize(var_vec.size() + num);
......@@ -119,7 +119,7 @@ void getBBoxVarFromPriorData(const T* prior_data, const size_t num,
}
}
template <typename T>
BBox<T> decodeBBoxWithVar(BBox<T>& prior_bbox,
BBox<T> decode_bbox_with_var(BBox<T>& prior_bbox,
const std::vector<T>& prior_bbox_var,
const std::vector<T>& loc_pred_data) {
T prior_bbox_width = prior_bbox.get_width();
......@@ -147,12 +147,12 @@ BBox<T> decodeBBoxWithVar(BBox<T>& prior_bbox,
return decoded_bbox;
}
template <typename T1, typename T2>
bool sortScorePairDescend(const std::pair<T1, T2>& pair1,
bool sort_score_pair_descend(const std::pair<T1, T2>& pair1,
const std::pair<T1, T2>& pair2) {
return pair1.first > pair2.first;
}
template <typename T>
T jaccardOverlap(const BBox<T>& bbox1, const BBox<T>& bbox2) {
T jaccard_overlap(const BBox<T>& bbox1, const BBox<T>& bbox2) {
if (bbox2.x_min > bbox1.x_max || bbox2.x_max < bbox1.x_min ||
bbox2.y_min > bbox1.y_max || bbox2.y_max < bbox1.y_min) {
return 0.0;
......@@ -174,10 +174,10 @@ T jaccardOverlap(const BBox<T>& bbox1, const BBox<T>& bbox2) {
}
template <typename T>
void applyNMSFast(const std::vector<BBox<T>>& bboxes, const T* conf_score_data,
size_t class_idx, size_t top_k, T conf_threshold,
T nms_threshold, size_t num_priors, size_t num_classes,
std::vector<size_t>* indices) {
void apply_nms_fast(const std::vector<BBox<T>>& bboxes,
const T* conf_score_data, size_t class_idx, size_t top_k,
T conf_threshold, T nms_threshold, size_t num_priors,
size_t num_classes, std::vector<size_t>* indices) {
std::vector<std::pair<T, size_t>> scores;
for (size_t i = 0; i < num_priors; ++i) {
size_t conf_offset = i * num_classes + class_idx;
......@@ -185,7 +185,7 @@ void applyNMSFast(const std::vector<BBox<T>>& bboxes, const T* conf_score_data,
scores.push_back(std::make_pair(conf_score_data[conf_offset], i));
}
std::stable_sort(scores.begin(), scores.end(),
sortScorePairDescend<T, size_t>);
sort_score_pair_descend<T, size_t>);
if (top_k > 0 && top_k < scores.size()) scores.resize(top_k);
while (scores.size() > 0) {
const size_t idx = scores.front().second;
......@@ -193,7 +193,7 @@ void applyNMSFast(const std::vector<BBox<T>>& bboxes, const T* conf_score_data,
for (size_t i = 0; i < indices->size(); ++i) {
if (keep) {
const size_t saved_idx = (*indices)[i];
T overlap = jaccardOverlap<T>(bboxes[idx], bboxes[saved_idx]);
T overlap = jaccard_overlap<T>(bboxes[idx], bboxes[saved_idx]);
keep = overlap <= nms_threshold;
} else {
break;
......@@ -204,7 +204,7 @@ void applyNMSFast(const std::vector<BBox<T>>& bboxes, const T* conf_score_data,
}
}
template <typename T>
int getDetectionIndices(
int get_detection_indices(
const T* conf_data, const size_t num_priors, const size_t num_classes,
const size_t background_label_id, const size_t batch_size,
const T conf_threshold, const size_t nms_top_k, const T nms_threshold,
......@@ -219,7 +219,7 @@ int getDetectionIndices(
size_t conf_offset = n * num_priors * num_classes;
for (size_t c = 0; c < num_classes; ++c) {
if (c == background_label_id) continue;
applyNMSFast<T>(decoded_bboxes, conf_data + conf_offset, c, nms_top_k,
apply_nms_fast<T>(decoded_bboxes, conf_data + conf_offset, c, nms_top_k,
conf_threshold, nms_threshold, num_priors, num_classes,
&(indices[c]));
num_detected += indices[c].size();
......@@ -237,7 +237,7 @@ int getDetectionIndices(
}
}
std::sort(score_index_pairs.begin(), score_index_pairs.end(),
sortScorePairDescend<T, std::pair<size_t, size_t>>);
sort_score_pair_descend<T, std::pair<size_t, size_t>>);
score_index_pairs.resize(top_k);
std::map<size_t, std::vector<size_t>> new_indices;
for (size_t i = 0; i < score_index_pairs.size(); ++i) {
......@@ -266,7 +266,7 @@ BBox<T> clipBBox(const BBox<T>& bbox) {
return clipped_bbox;
}
template <typename T>
void getDetectionOutput(
void get_detection_output(
const T* conf_data, const size_t num_kept, const size_t num_priors,
const size_t num_classes, const size_t batch_size,
const std::vector<std::map<size_t, std::vector<size_t>>>& all_indices,
......
Markdown is supported
0% 或 .
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 或