提交 144016fc 编写于 作者: D dengkaipeng

fix adaptive_pool and yolov3_loss. test=develop

上级 eb65b4e4
......@@ -144,30 +144,36 @@ class Yolov3LossOpMaker : public framework::OpProtoAndCheckerMaker {
"The ignore threshold to ignore confidence loss.")
.SetDefault(0.7);
AddComment(R"DOC(
This operator generate yolov3 loss by given predict result and ground
This operator generates yolov3 loss based on given predict result and ground
truth boxes.
The output of previous network is in shape [N, C, H, W], while H and W
should be the same, specify the grid size, each grid point predict given
number boxes, this given number is specified by anchors, it should be
half anchors length, which following will be represented as S. In the
second dimention(the channel dimention), C should be S * (class_num + 5),
class_num is the box categoriy number of source dataset(such as coco),
so in the second dimention, stores 4 box location coordinates x, y, w, h
and confidence score of the box and class one-hot key of each anchor box.
should be the same, H and W specify the grid size, each grid point predict
given number boxes, this given number, which following will be represented as S,
is specified by the number of anchors, In the second dimension(the channel
dimension), C should be equal to S * (class_num + 5), class_num is the object
category number of source dataset(such as 80 in coco dataset), so in the
second(channel) dimension, apart from 4 box location coordinates x, y, w, h,
also includes confidence score of the box and class one-hot key of each anchor box.
While the 4 location coordinates if $$tx, ty, tw, th$$, the box predictions
correspnd to:
Assume the 4 location coordinates is :math:`t_x, t_y, t_w, t_h`, the box predictions
should be following:
$$
b_x = \sigma(t_x) + c_x
b_y = \sigma(t_y) + c_y
b_x = \\sigma(t_x) + c_x
$$
$$
b_y = \\sigma(t_y) + c_y
$$
$$
b_w = p_w e^{t_w}
$$
$$
b_h = p_h e^{t_h}
$$
While $$c_x, c_y$$ is the left top corner of current grid and $$p_w, p_h$$
is specified by anchors.
In the equaltion above, :math:`c_x, c_y` is the left top corner of current grid
and :math:`p_w, p_h` is specified by anchors.
As for confidence score, it is the logistic regression value of IoU between
anchor boxes and ground truth boxes, the score of the anchor box which has
......
......@@ -260,34 +260,39 @@ Example:
$$
For exclusive = false:
.. math::
hstart &= i * strides[0] - paddings[0] \\
hend &= hstart + ksize[0] \\
wstart &= j * strides[1] - paddings[1] \\
wend &= wstart + ksize[1] \\
Output(i ,j) &= \frac{sum(Input[hstart:hend, wstart:wend])}{ksize[0] * ksize[1]}
$$
hstart = i * strides[0] - paddings[0]
$$
$$
hend = hstart + ksize[0]
$$
$$
wstart = j * strides[1] - paddings[1]
$$
$$
wend = wstart + ksize[1]
$$
$$
Output(i ,j) = \\frac{sum(Input[hstart:hend, wstart:wend])}{ksize[0] * ksize[1]}
$$
For exclusive = true:
$$
hstart = max(0, i * strides[0] - paddings[0])
$$
$$
hend = min(H, hstart + ksize[0])
$$
$$
wstart = max(0, j * strides[1] - paddings[1])
$$
$$
wend = min(W, wstart + ksize[1])
$$
$$
Output(i ,j) = \\frac{sum(Input[hstart:hend, wstart:wend])}{(hend - hstart) * (wend - wstart)}
$$
.. math::
hstart &= max(0, i * strides[0] - paddings[0]) \\
hend &= min(H, hstart + ksize[0]) \\
wstart &= max(0, j * strides[1] - paddings[1]) \\
wend &= min(W, wstart + ksize[1]) \\
Output(i ,j) &= \frac{sum(Input[hstart:hend, wstart:wend])}{(hend - hstart) * (wend - wstart)}
For adaptive = true:
.. math::
hstart &= floor(i * H_{in} / H_{out}) \\
hend &= ceil((i + 1) * H_{in} / H_{out}) \\
wstart &= floor(j * W_{in} / W_{out}) \\
wend &= ceil((j + 1) * W_{in} / W_{out}) \\
Output(i ,j) &= \frac{sum(Input[hstart:hend, wstart:wend])}{(hend - hstart) * (wend - wstart)}
)DOC");
}
......@@ -417,39 +422,47 @@ Example:
$$
For exclusive = false:
.. math::
dstart &= i * strides[0] - paddings[0] \\
dend &= dstart + ksize[0] \\
hstart &= j * strides[1] - paddings[1] \\
hend &= hstart + ksize[1] \\
wstart &= k * strides[2] - paddings[2] \\
wend &= wstart + ksize[2] \\
Output(i ,j, k) &= \frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{ksize[0] * ksize[1] * ksize[2]}
$$
dstart = i * strides[0] - paddings[0]
$$
$$
dend = dstart + ksize[0]
$$
$$
hstart = j * strides[1] - paddings[1]
$$
$$
hend = hstart + ksize[1]
$$
$$
wstart = k * strides[2] - paddings[2]
$$
$$
wend = wstart + ksize[2]
$$
$$
Output(i ,j, k) = \\frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{ksize[0] * ksize[1] * ksize[2]}
$$
For exclusive = true:
.. math::
dstart &= max(0, i * strides[0] - paddings[0]) \\
dend &= min(D, dstart + ksize[0]) \\
hend &= min(H, hstart + ksize[1]) \\
wstart &= max(0, k * strides[2] - paddings[2]) \\
wend &= min(W, wstart + ksize[2]) \\
Output(i ,j, k) &= \frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{(dend - dstart) * (hend - hstart) * (wend - wstart)}
For adaptive = true:
.. math::
dstart &= floor(i * D_{in} / D_{out}) \\
dend &= ceil((i + 1) * D_{in} / D_{out}) \\
hstart &= floor(j * H_{in} / H_{out}) \\
hend &= ceil((j + 1) * H_{in} / H_{out}) \\
wstart &= floor(k * W_{in} / W_{out}) \\
wend &= ceil((k + 1) * W_{in} / W_{out}) \\
Output(i ,j, k) &= \frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{(dend - dstart) * (hend - hstart) * (wend - wstart)}
$$
dstart = max(0, i * strides[0] - paddings[0])
$$
$$
dend = min(D, dstart + ksize[0])
$$
$$
hend = min(H, hstart + ksize[1])
$$
$$
wstart = max(0, k * strides[2] - paddings[2])
$$
$$
wend = min(W, wstart + ksize[2])
$$
$$
Output(i ,j, k) = \\frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{(dend - dstart) * (hend - hstart) * (wend - wstart)}
$$
)DOC");
}
......
......@@ -545,15 +545,16 @@ def yolov3_loss(x,
TypeError: Attr ignore_thresh of yolov3_loss must be a float number
Examples:
.. code-block:: python
x = fluid.layers.data(name='x', shape=[255, 13, 13], dtype='float32')
gtbox = fluid.layers.data(name='gtbox', shape=[6, 5], dtype='float32')
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 = [0, 1, 2]
loss = fluid.layers.yolov3_loss(x=x, gtbox=gtbox, class_num=80, anchors=anchors,
ignore_thresh=0.5, downsample_ratio=32)
.. code-block:: python
x = fluid.layers.data(name='x', shape=[255, 13, 13], dtype='float32')
gtbox = fluid.layers.data(name='gtbox', shape=[6, 5], dtype='float32')
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]
anchor_mask = [0, 1, 2]
loss = fluid.layers.yolov3_loss(x=x, gtbox=gtbox, gtlabel=gtlabel, anchors=anchors,
anchor_mask=anchor_mask, class_num=80,
ignore_thresh=0.7, downsample_ratio=32)
"""
helper = LayerHelper('yolov3_loss', **locals())
......
......@@ -2577,6 +2577,20 @@ def adaptive_pool2d(input,
represent height and width, respectively. Also the H and W dimensions of output(Out)
is same as Parameter(pool_size).
For average adaptive pool2d:
.. math::
hstart &= floor(i * H_{in} / H_{out})
hend &= ceil((i + 1) * H_{in} / H_{out})
wstart &= floor(j * W_{in} / W_{out})
wend &= ceil((j + 1) * W_{in} / W_{out})
Output(i ,j) &= \\frac{sum(Input[hstart:hend, wstart:wend])}{(hend - hstart) * (wend - wstart)}
Args:
input (Variable): The input tensor of pooling operator. The format of
input tensor is NCHW, where N is batch size, C is
......@@ -2675,6 +2689,24 @@ def adaptive_pool3d(input,
three elements which represent height and width, respectively. Also the D, H and W
dimensions of output(Out) is same as Parameter(pool_size).
For average adaptive pool3d:
.. math::
dstart &= floor(i * D_{in} / D_{out})
dend &= ceil((i + 1) * D_{in} / D_{out})
hstart &= floor(j * H_{in} / H_{out})
hend &= ceil((j + 1) * H_{in} / H_{out})
wstart &= floor(k * W_{in} / W_{out})
wend &= ceil((k + 1) * W_{in} / W_{out})
Output(i ,j, k) &= \\frac{sum(Input[dstart:dend, hstart:hend, wstart:wend])}{(dend - dstart) * (hend - hstart) * (wend - wstart)}
Args:
input (Variable): The input tensor of pooling operator. The format of
input tensor is NCDHW, where N is batch size, C is
......
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