m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

downsample -> downsample_ratio. test=develop

Browse Source
revert-15296-async_double_buffered_py_reader
dengkaipeng 6 years ago
parent ae0b0d5f93
commit 733bb82ec0
4 changed files with 29 additions and 26 deletions
Show Stats Download Patch File Download Diff File

2
paddle/fluid/API.spec
Unescape View File

@ -324,7 +324,7 @@ paddle.fluid.layers.generate_mask_labels ArgSpec(args=['im_info', 'gt_classes',
paddle.fluid.layers.iou_similarity ArgSpec(args=['x', 'y', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.iou_similarity ArgSpec(args=['x', 'y', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.box_coder ArgSpec(args=['prior_box', 'prior_box_var', 'target_box', 'code_type', 'box_normalized', 'name'], varargs=None, keywords=None, defaults=('encode_center_size', True, None)) paddle.fluid.layers.box_coder ArgSpec(args=['prior_box', 'prior_box_var', 'target_box', 'code_type', 'box_normalized', 'name'], varargs=None, keywords=None, defaults=('encode_center_size', True, None))
paddle.fluid.layers.polygon_box_transform ArgSpec(args=['input', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.polygon_box_transform ArgSpec(args=['input', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.yolov3_loss ArgSpec(args=['x', 'gtbox', 'gtlabel', 'anchors', 'anchor_mask', 'class_num', 'ignore_thresh', 'downsample', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.yolov3_loss ArgSpec(args=['x', 'gtbox', 'gtlabel', 'anchors', 'anchor_mask', 'class_num', 'ignore_thresh', 'downsample_ratio', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.multiclass_nms ArgSpec(args=['bboxes', 'scores', 'score_threshold', 'nms_top_k', 'keep_top_k', 'nms_threshold', 'normalized', 'nms_eta', 'background_label', 'name'], varargs=None, keywords=None, defaults=(0.3, True, 1.0, 0, None)) paddle.fluid.layers.multiclass_nms ArgSpec(args=['bboxes', 'scores', 'score_threshold', 'nms_top_k', 'keep_top_k', 'nms_threshold', 'normalized', 'nms_eta', 'background_label', 'name'], varargs=None, keywords=None, defaults=(0.3, True, 1.0, 0, None))
paddle.fluid.layers.accuracy ArgSpec(args=['input', 'label', 'k', 'correct', 'total'], varargs=None, keywords=None, defaults=(1, None, None)) paddle.fluid.layers.accuracy ArgSpec(args=['input', 'label', 'k', 'correct', 'total'], varargs=None, keywords=None, defaults=(1, None, None))
paddle.fluid.layers.auc ArgSpec(args=['input', 'label', 'curve', 'num_thresholds', 'topk', 'slide_steps'], varargs=None, keywords=None, defaults=('ROC', 4095, 1, 1)) paddle.fluid.layers.auc ArgSpec(args=['input', 'label', 'curve', 'num_thresholds', 'topk', 'slide_steps'], varargs=None, keywords=None, defaults=('ROC', 4095, 1, 1))

2
paddle/fluid/operators/yolov3_loss_op.cc
Unescape View File

@ -135,7 +135,7 @@ class Yolov3LossOpMaker : public framework::OpProtoAndCheckerMaker {
"The mask index of anchors used in " "The mask index of anchors used in "
"current YOLOv3 loss calculation.") "current YOLOv3 loss calculation.")
.SetDefault(std::vector<int>{}); .SetDefault(std::vector<int>{});
AddAttr<int>("downsample", AddAttr<int>("downsample_ratio",
"The downsample ratio from network input to YOLOv3 loss " "The downsample ratio from network input to YOLOv3 loss "
"input, so 32, 16, 8 should be set for the first, second, " "input, so 32, 16, 8 should be set for the first, second, "
"and thrid YOLOv3 loss operators.") "and thrid YOLOv3 loss operators.")

41
paddle/fluid/operators/yolov3_loss_op.h
Unescape View File

@ -32,7 +32,7 @@ static inline bool LessEqualZero(T x) {
} }
template <typename T> template <typename T>
static T SCE(T x, T label) { static T SigmoidCrossEntropy(T x, T label) {
return (x > 0 ? x : 0.0) - x * label + std::log(1.0 + std::exp(-std::abs(x))); return (x > 0 ? x : 0.0) - x * label + std::log(1.0 + std::exp(-std::abs(x)));
} }
@ -42,7 +42,7 @@ static T L2Loss(T x, T y) {
} }
template <typename T> template <typename T>
static T SCEGrad(T x, T label) { static T SigmoidCrossEntropyGrad(T x, T label) {
return 1.0 / (1.0 + std::exp(-x)) - label; return 1.0 / (1.0 + std::exp(-x)) - label;
} }
@ -62,7 +62,7 @@ static int GetMaskIndex(std::vector<int> mask, int val) {
template <typename T> template <typename T>
struct Box { struct Box {
float x, y, w, h; T x, y, w, h;
}; };
template <typename T> template <typename T>
@ -128,8 +128,8 @@ static void CalcBoxLocationLoss(T* loss, const T* input, Box<T> gt,
T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]); T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]);
T scale = (2.0 - gt.w * gt.h); T scale = (2.0 - gt.w * gt.h);
loss[0] += SCE<T>(input[box_idx], tx) * scale; loss[0] += SigmoidCrossEntropy<T>(input[box_idx], tx) * scale;
loss[0] += SCE<T>(input[box_idx + stride], ty) * scale; loss[0] += SigmoidCrossEntropy<T>(input[box_idx + stride], ty) * scale;
loss[0] += L2Loss<T>(input[box_idx + 2 * stride], tw) * scale; loss[0] += L2Loss<T>(input[box_idx + 2 * stride], tw) * scale;
loss[0] += L2Loss<T>(input[box_idx + 3 * stride], th) * scale; loss[0] += L2Loss<T>(input[box_idx + 3 * stride], th) * scale;
} }
@ -145,9 +145,10 @@ static void CalcBoxLocationLossGrad(T* input_grad, const T loss, const T* input,
T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]); T th = std::log(gt.h * input_size / anchors[2 * an_idx + 1]);
T scale = (2.0 - gt.w * gt.h); T scale = (2.0 - gt.w * gt.h);
input_grad[box_idx] = SCEGrad<T>(input[box_idx], tx) * scale * loss; input_grad[box_idx] =
SigmoidCrossEntropyGrad<T>(input[box_idx], tx) * scale * loss;
input_grad[box_idx + stride] = input_grad[box_idx + stride] =
SCEGrad<T>(input[box_idx + stride], ty) * scale * loss; SigmoidCrossEntropyGrad<T>(input[box_idx + stride], ty) * scale * loss;
input_grad[box_idx + 2 * stride] = input_grad[box_idx + 2 * stride] =
L2LossGrad<T>(input[box_idx + 2 * stride], tw) * scale * loss; L2LossGrad<T>(input[box_idx + 2 * stride], tw) * scale * loss;
input_grad[box_idx + 3 * stride] = input_grad[box_idx + 3 * stride] =
@ -160,7 +161,7 @@ static inline void CalcLabelLoss(T* loss, const T* input, const int index,
const int stride) { const int stride) {
for (int i = 0; i < class_num; i++) { for (int i = 0; i < class_num; i++) {
T pred = input[index + i * stride]; T pred = input[index + i * stride];
loss[0] += SCE<T>(pred, (i == label) ? 1.0 : 0.0); loss[0] += SigmoidCrossEntropy<T>(pred, (i == label) ? 1.0 : 0.0);
} }
} }
@ -172,7 +173,7 @@ static inline void CalcLabelLossGrad(T* input_grad, const T loss,
for (int i = 0; i < class_num; i++) { for (int i = 0; i < class_num; i++) {
T pred = input[index + i * stride]; T pred = input[index + i * stride];
input_grad[index + i * stride] = input_grad[index + i * stride] =
SCEGrad<T>(pred, (i == label) ? 1.0 : 0.0) * loss; SigmoidCrossEntropyGrad<T>(pred, (i == label) ? 1.0 : 0.0) * loss;
} }
} }
@ -187,11 +188,11 @@ static inline void CalcObjnessLoss(T* loss, const T* input, const T* objness,
for (int l = 0; l < w; l++) { for (int l = 0; l < w; l++) {
T obj = objness[k * w + l]; T obj = objness[k * w + l];
if (obj > 1e-5) { if (obj > 1e-5) {
// positive sample: obj = mixup score // positive sample: obj = 1
loss[i] += SCE<T>(input[k * w + l], 1.0); loss[i] += SigmoidCrossEntropy<T>(input[k * w + l], 1.0);
} else if (obj > -0.5) { } else if (obj > -0.5) {
// negetive sample: obj = 0 // negetive sample: obj = 0
loss[i] += SCE<T>(input[k * w + l], 0.0); loss[i] += SigmoidCrossEntropy<T>(input[k * w + l], 0.0);
} }
} }
} }
@ -213,9 +214,11 @@ static inline void CalcObjnessLossGrad(T* input_grad, const T* loss,
for (int l = 0; l < w; l++) { for (int l = 0; l < w; l++) {
T obj = objness[k * w + l]; T obj = objness[k * w + l];
if (obj > 1e-5) { if (obj > 1e-5) {
input_grad[k * w + l] = SCEGrad<T>(input[k * w + l], 1.0) * loss[i]; input_grad[k * w + l] =
SigmoidCrossEntropyGrad<T>(input[k * w + l], 1.0) * loss[i];
} else if (obj > -0.5) { } else if (obj > -0.5) {
input_grad[k * w + l] = SCEGrad<T>(input[k * w + l], 0.0) * loss[i]; input_grad[k * w + l] =
SigmoidCrossEntropyGrad<T>(input[k * w + l], 0.0) * loss[i];
} }
} }
} }
@ -256,7 +259,7 @@ class Yolov3LossKernel : public framework::OpKernel<T> {
auto anchor_mask = ctx.Attr<std::vector<int>>("anchor_mask"); auto anchor_mask = ctx.Attr<std::vector<int>>("anchor_mask");
int class_num = ctx.Attr<int>("class_num"); int class_num = ctx.Attr<int>("class_num");
float ignore_thresh = ctx.Attr<float>("ignore_thresh"); float ignore_thresh = ctx.Attr<float>("ignore_thresh");
int downsample = ctx.Attr<int>("downsample"); int downsample_ratio = ctx.Attr<int>("downsample_ratio");
const int n = input->dims()[0]; const int n = input->dims()[0];
const int h = input->dims()[2]; const int h = input->dims()[2];
@ -264,7 +267,7 @@ class Yolov3LossKernel : public framework::OpKernel<T> {
const int an_num = anchors.size() / 2; const int an_num = anchors.size() / 2;
const int mask_num = anchor_mask.size(); const int mask_num = anchor_mask.size();
const int b = gt_box->dims()[1]; const int b = gt_box->dims()[1];
int input_size = downsample * h; int input_size = downsample_ratio * h;
const int stride = h * w; const int stride = h * w;
const int an_stride = (class_num + 5) * stride; const int an_stride = (class_num + 5) * stride;
@ -308,7 +311,7 @@ class Yolov3LossKernel : public framework::OpKernel<T> {
} }
} }
// If best IoU is greater then ignore_thresh, // If best IoU is bigger then ignore_thresh,
// ignore the objectness loss. // ignore the objectness loss.
if (best_iou > ignore_thresh) { if (best_iou > ignore_thresh) {
int obj_idx = (i * mask_num + j) * stride + k * w + l; int obj_idx = (i * mask_num + j) * stride + k * w + l;
@ -388,7 +391,7 @@ class Yolov3LossGradKernel : public framework::OpKernel<T> {
auto anchors = ctx.Attr<std::vector<int>>("anchors"); auto anchors = ctx.Attr<std::vector<int>>("anchors");
auto anchor_mask = ctx.Attr<std::vector<int>>("anchor_mask"); auto anchor_mask = ctx.Attr<std::vector<int>>("anchor_mask");
int class_num = ctx.Attr<int>("class_num"); int class_num = ctx.Attr<int>("class_num");
int downsample = ctx.Attr<int>("downsample"); int downsample_ratio = ctx.Attr<int>("downsample_ratio");
const int n = input_grad->dims()[0]; const int n = input_grad->dims()[0];
const int c = input_grad->dims()[1]; const int c = input_grad->dims()[1];
@ -396,7 +399,7 @@ class Yolov3LossGradKernel : public framework::OpKernel<T> {
const int w = input_grad->dims()[3]; const int w = input_grad->dims()[3];
const int mask_num = anchor_mask.size(); const int mask_num = anchor_mask.size();
const int b = gt_match_mask->dims()[1]; const int b = gt_match_mask->dims()[1];
int input_size = downsample * h; int input_size = downsample_ratio * h;
const int stride = h * w; const int stride = h * w;
const int an_stride = (class_num + 5) * stride; const int an_stride = (class_num + 5) * stride;

10
python/paddle/fluid/layers/detection.py
Unescape View File

@ -416,7 +416,7 @@ def yolov3_loss(x,
anchor_mask, anchor_mask,
class_num, class_num,
ignore_thresh, ignore_thresh,
downsample, downsample_ratio,
name=None): name=None):
""" """
${comment} ${comment}
@ -434,7 +434,7 @@ def yolov3_loss(x,
anchor_mask (list|tuple): ${anchor_mask_comment} anchor_mask (list|tuple): ${anchor_mask_comment}
class_num (int): ${class_num_comment} class_num (int): ${class_num_comment}
ignore_thresh (float): ${ignore_thresh_comment} ignore_thresh (float): ${ignore_thresh_comment}
downsample (int): ${downsample_comment} downsample_ratio (int): ${downsample_ratio_comment}
name (string): the name of yolov3 loss name (string): the name of yolov3 loss
Returns: Returns:
@ -456,8 +456,8 @@ def yolov3_loss(x,
gtlabel = fluid.layers.data(name='gtlabel', shape=[6, 1], dtype='int32') gtlabel = fluid.layers.data(name='gtlabel', shape=[6, 1], dtype='int32')
anchors = [10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326] anchors = [10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326]
anchors = [0, 1, 2] anchors = [0, 1, 2]
loss = fluid.layers.yolov3_loss(x=x, gtbox=gtbox, class_num=80 loss = fluid.layers.yolov3_loss(x=x, gtbox=gtbox, class_num=80, anchors=anchors,
anchors=anchors, ignore_thresh=0.5) ignore_thresh=0.5, downsample_ratio=32)
""" """
helper = LayerHelper('yolov3_loss', **locals()) helper = LayerHelper('yolov3_loss', **locals())
@ -491,7 +491,7 @@ def yolov3_loss(x,
"anchor_mask": anchor_mask, "anchor_mask": anchor_mask,
"class_num": class_num, "class_num": class_num,
"ignore_thresh": ignore_thresh, "ignore_thresh": ignore_thresh,
"downsample": downsample, "downsample_ratio": downsample_ratio,
} }
helper.append_op( helper.append_op(

Write Preview
Loading…
Cancel
Save