caffe 参数介绍
solver.prototxt
<code class="hljs http has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-attribute" style="box-sizing: border-box;">net</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"models/bvlc_alexnet/train_val.prototxt" </span>
<span class="hljs-attribute" style="box-sizing: border-box;">test_iter</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">1000 # </span>
<span class="hljs-attribute" style="box-sizing: border-box;">test_interval</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">1000 # </span>
<span class="hljs-attribute" style="box-sizing: border-box;">base_lr</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">0.01 # 开始的学习率</span>
<span class="hljs-attribute" style="box-sizing: border-box;">lr_policy</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"step" # 学习率的drop是以gamma在每一次迭代中</span>
<span class="hljs-attribute" style="box-sizing: border-box;">gamma</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">0.1</span>
<span class="hljs-attribute" style="box-sizing: border-box;">stepsize</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">100000 # 每stepsize的迭代降低学习率:乘以gamma</span>
<span class="hljs-attribute" style="box-sizing: border-box;">display</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">20 # 没display次打印显示loss</span>
<span class="hljs-attribute" style="box-sizing: border-box;">max_iter</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">450000 # train 最大迭代max_iter </span>
<span class="hljs-attribute" style="box-sizing: border-box;">momentum</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">0.9 #</span>
<span class="hljs-attribute" style="box-sizing: border-box;">weight_decay</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">0.0005 #</span>
<span class="hljs-attribute" style="box-sizing: border-box;">snapshot</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">10000 # 没迭代snapshot次,保存一次快照</span>
<span class="hljs-attribute" style="box-sizing: border-box;">snapshot_prefix</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;"> "models/bvlc_reference_caffenet/caffenet_train"</span>
<span class="hljs-attribute" style="box-sizing: border-box;">solver_mode</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">GPU # 使用的模式是GPU </span></code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li></ul>
-
test_iter
在测试的时候,需要迭代的次数,即test_iter* batchsize(测试集的)=测试集的大小,测试集的 batchsize可以在prototx文件里设置。
-
test_interval
训练的时候,每迭代test_interval次就进行一次测试。
-
momentum
灵感来自于牛顿第一定律,基本思路是为寻优加入了“惯性”的影响,这样一来,当误差曲面中存在平坦区的时候,SGD可以更快的速度学习。
wi←m∗wi−η∂E∂wi
train_val.prototxt
<code class="hljs bash has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">layer { <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 数据层</span>
name: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"data"</span>
<span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">type</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"Data"</span>
top: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"data"</span>
top: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"label"</span>
include {
phase: TRAIN <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 表明这是在训练阶段才包括进去</span>
}
transform_param { <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 对数据进行预处理</span>
mirror: <span class="hljs-literal" style="color: rgb(0, 102, 102); box-sizing: border-box;">true</span> <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 是否做镜像</span>
crop_size: <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">227</span>
<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 减去均值文件</span>
mean_file: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"data/ilsvrc12/imagenet_mean.binaryproto"</span>
}
data_param { <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 设定数据的来源</span>
<span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">source</span>: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"examples/imagenet/ilsvrc12_train_lmdb"</span>
batch_size: <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>
backend: LMDB
}
}</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li></ul>
<code class="hljs css has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-tag" style="color: rgb(0, 0, 0); box-sizing: border-box;">layer</span> <span class="hljs-rules" style="box-sizing: border-box;">{
<span class="hljs-rule" style="box-sizing: border-box;"><span class="hljs-attribute" style="box-sizing: border-box;">name</span>:<span class="hljs-value" style="box-sizing: border-box; color: rgb(0, 102, 102);"> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"data"</span>
type: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"Data"</span>
top: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"data"</span>
top: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"label"</span>
include {
phase: TEST # 测试阶段
</span></span></span>}
<span class="hljs-tag" style="color: rgb(0, 0, 0); box-sizing: border-box;">transform_param</span> <span class="hljs-rules" style="box-sizing: border-box;">{
<span class="hljs-rule" style="box-sizing: border-box;"><span class="hljs-attribute" style="box-sizing: border-box;">mirror</span>:<span class="hljs-value" style="box-sizing: border-box; color: rgb(0, 102, 102);"> false # 是否做镜像
crop_size: <span class="hljs-number" style="box-sizing: border-box;">227</span>
# 减去均值文件
mean_file: <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"data/ilsvrc12/imagenet_mean.binaryproto"</span>
</span></span></span>}
<span class="hljs-tag" style="color: rgb(0, 0, 0); box-sizing: border-box;">data_param</span> <span class="hljs-rules" style="box-sizing: border-box;">{
<span class="hljs-rule" style="box-sizing: border-box;"><span class="hljs-attribute" style="box-sizing: border-box;">source</span>:<span class="hljs-value" style="box-sizing: border-box; color: rgb(0, 102, 102);"> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"examples/imagenet/ilsvrc12_val_lmdb"</span>
batch_size: <span class="hljs-number" style="box-sizing: border-box;">50</span>
backend: LMDB
</span></span></span>}
}</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li></ul>
-
lr_mult
学习率,但是最终的学习率需要乘以 solver.prototxt 配置文件中的 base_lr .
如果有两个 lr_mult, 则第一个表示 weight 的学习率,第二个表示 bias 的学习率
一般 bias 的学习率是 weight 学习率的2倍’
-
decay_mult
权值衰减,为了避免模型的over-fitting,需要对cost function加入规范项。
wi←wi−η∂E∂wi−ηλwi
-
num_output
卷积核(filter)的个数
-
kernel_size
卷积核的大小。
如果卷积核的长和宽不等,需要用 kernel_h 和 kernel_w 分别设定
-
stride
卷积核的步长,默认为1。也可以用stride_h和stride_w来设置。
-
pad
扩充边缘,默认为0,不扩充。
扩充的时候是左右、上下对称的,比如卷积核的大小为5*5,那么pad设置为2,则四个边缘都扩充2个像素,即宽度和高度都扩充了4个像素,这样卷积运算之后的特征图就不会变小。
也可以通过pad_h和pad_w来分别设定。
-
weight_filler
权值初始化。 默认为“constant”,值全为0.
很多时候我们用”xavier”算法来进行初始化,也可以设置为”gaussian”
<code class="hljs css has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-tag" style="color: rgb(0, 0, 0); box-sizing: border-box;">weight_filler</span> <span class="hljs-rules" style="box-sizing: border-box;">{
<span class="hljs-rule" style="box-sizing: border-box;"><span class="hljs-attribute" style="box-sizing: border-box;">type</span>:<span class="hljs-value" style="box-sizing: border-box; color: rgb(0, 102, 102);"> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"gaussian"</span>
std: <span class="hljs-number" style="box-sizing: border-box;">0.01</span>
</span></span></span>}</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul>
偏置项的初始化。一般设置为”constant”, 值全为0。
<code class="hljs css has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-tag" style="color: rgb(0, 0, 0); box-sizing: border-box;">bias_filler</span> <span class="hljs-rules" style="box-sizing: border-box;">{
<span class="hljs-rule" style="box-sizing: border-box;"><span class="hljs-attribute" style="box-sizing: border-box;">type</span>:<span class="hljs-value" style="box-sizing: border-box; color: rgb(0, 102, 102);"> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"constant"</span>
value: <span class="hljs-number" style="box-sizing: border-box;">0</span>
</span></span></span>}</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul>
-
bias_term
是否开启偏置项,默认为true, 开启
-
group
分组,默认为1组。如果大于1,我们限制卷积的连接操作在一个子集内。
卷积分组可以减少网络的参数,至于是否还有其他的作用就不清楚了。
每个input是需要和每一个kernel都进行连接的,但是由于分组的原因其只是与部分的kernel进行连接的
如: 我们根据图像的通道来分组,那么第i个输出分组只能与第i个输入分组进行连接。
-
pool
池化方法,默认为MAX。目前可用的方法有 MAX, AVE, 或 STOCHASTIC
-
dropout_ratio
丢弃数据的概率
(以下 caffe旧版 来源于http://demo.netfoucs.com/danieljianfeng/article/details/42929283)
1. Vision Layers
1.1 卷积层(Convolution)
类型:CONVOLUTION
例子
layers {
name: "conv1"
type: CONVOLUTION
bottom: "data"
top: "conv1"
blobs_lr: 1 <span style="font-family: 'microsoft yahei';"># learning rate multiplier for the filters </span>
<span style="font-family: 'microsoft yahei';"> blobs_lr: 2 # learning rate multiplier for the biases </span>
<span style="font-family: 'microsoft yahei';"> weight_decay: 1 # weight decay multiplier for the filters</span>
<span style="font-family: 'microsoft yahei';"> weight_decay: 0 # weight decay multiplier for the biases </span>
<span style="font-family: 'microsoft yahei';">convolution_param { num_output: 96 # learn 96 filters </span>
<span style="font-family: 'microsoft yahei';"> kernel_size: 11 # each filter is 11x11 </span>
<span style="font-family: 'microsoft yahei';"> stride: 4 # step 4 pixels between each filter application </span>
<span style="font-family: 'microsoft yahei';">weight_filler { </span>
<span style="font-family: 'microsoft yahei';"> type: "gaussian" # initialize the filters from a Gaussian </span>
<span style="font-family: 'microsoft yahei';"> std: 0.01 # distribution with stdev 0.01 (default mean: 0) </span>
<span style="font-family: 'microsoft yahei';"> } </span>
<span style="font-family: 'microsoft yahei';"> bias_filler { </span>
<span style="font-family: 'microsoft yahei';"> type: "constant" # initialize the biases to zero (0) </span>
<span style="font-family: 'microsoft yahei';"> value: 0 } </span>
<span style="font-family: 'microsoft yahei';">}</span>
<span style="font-family: 'microsoft yahei';">}</span>
blobs_lr: 学习率调整的参数,在上面的例子中设置权重学习率和运行中求解器给出的学习率一样,同时是偏置学习率为权重的两倍。
weight_decay:
卷积层的重要参数
必须参数:
num_output (c_o):过滤器的个数
kernel_size (or kernel_h and kernel_w):过滤器的大小
可选参数:
weight_filler [default type: 'constant' value: 0]:参数的初始化方法
bias_filler:偏置的初始化方法
bias_term [default true]:指定是否是否开启偏置项
pad (or pad_h and pad_w) [default 0]:指定在输入的每一边加上多少个像素
stride (or stride_h and stride_w) [default 1]:指定过滤器的步长
group (g) [default 1]: If g > 1, we restrict the connectivityof each filter to a subset of the input. Specifically, the input and outputchannels are separated into g groups, and the ith output group channels will beonly connected to the ith input group channels.
通过卷积后的大小变化:
输入:n * c_i * h_i * w_i
输出:n * c_o * h_o * w_o,其中h_o = (h_i + 2 * pad_h - kernel_h) /stride_h + 1,w_o通过同样的方法计算。
1.2 池化层(Pooling)
类型:POOLING
例子
layers { name: "pool1" type: POOLING bottom: "conv1" top: "pool1" pooling_param { pool: MAX kernel_size: 3 # pool over a 3x3 region stride: 2 # step two pixels (in the bottom blob) between pooling regions }}
卷积层的重要参数
必需参数:
kernel_size (or kernel_h and kernel_w):过滤器的大小
可选参数:
pool [default MAX]:pooling的方法,目前有MAX, AVE, 和STOCHASTIC三种方法
pad (or pad_h and pad_w) [default 0]:指定在输入的每一遍加上多少个像素
stride (or stride_h and stride_w) [default1]:指定过滤器的步长
通过池化后的大小变化:
输入:n * c_i * h_i * w_i
输出:n * c_o * h_o * w_o,其中h_o = (h_i + 2 * pad_h - kernel_h) /stride_h + 1,w_o通过同样的方法计算。
1.3 Local Response Normalization (LRN)
类型:LRN
Local ResponseNormalization是对一个局部的输入区域进行的归一化(激活a被加一个归一化权重(分母部分)生成了新的激活b),有两种不同的形式,一种的输入区域为相邻的channels(cross channel LRN),另一种是为同一个channel内的空间区域(within channel LRN)
计算公式:对每一个输入除以![](https://img-blog.csdn.net/20150120213228813?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvZGFuaWVsamlhbmZlbmc=/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center)
可选参数:
local_size [default 5]:对于cross channel LRN为需要求和的邻近channel的数量;对于within channel LRN为需要求和的空间区域的边长
alpha [default 1]:scaling参数
beta [default 5]:指数
norm_region [default ACROSS_CHANNELS]: 选择哪种LRN的方法ACROSS_CHANNELS 或者WITHIN_CHANNEL
2. Loss Layers
深度学习是通过最小化输出和目标的Loss来驱动学习。
2.1 Softmax
类型: SOFTMAX_LOSS
2.2 Sum-of-Squares / Euclidean
类型: EUCLIDEAN_LOSS
2.3 Hinge / Margin
类型: HINGE_LOSS
例子:
# L1 Normlayers { name: "loss" type: HINGE_LOSS bottom: "pred" bottom: "label"}# L2 Normlayers { name: "loss" type: HINGE_LOSS bottom: "pred" bottom: "label" top: "loss" hinge_loss_param { norm: L2 }}
可选参数:
norm [default L1]: 选择L1或者 L2范数
输入:
n * c * h * wPredictions
n * 1 * 1 * 1Labels
输出
1 * 1 * 1 * 1Computed Loss
2.4 Sigmoid Cross-Entropy
类型:SIGMOID_CROSS_ENTROPY_LOSS
2.5 Infogain
类型:INFOGAIN_LOSS
2.6 Accuracy and Top-k
类型:ACCURACY
用来计算输出和目标的正确率,事实上这不是一个loss,而且没有backward这一步。
3. 激励层(Activation / Neuron Layers)
一般来说,激励层是element-wise的操作,输入和输出的大小相同,一般情况下就是一个非线性函数。
3.1 ReLU / Rectified-Linear and Leaky-ReLU
类型: RELU
例子:
layers { name: "relu1" type: RELU bottom: "conv1" top: "conv1"}
可选参数:
negative_slope [default 0]:指定输入值小于零时的输出。
ReLU是目前使用做多的激励函数,主要因为其收敛更快,并且能保持同样效果。
标准的ReLU函数为max(x, 0),而一般为当x > 0时输出x,但x <= 0时输出negative_slope。RELU层支持in-place计算,这意味着bottom的输出和输入相同以避免内存的消耗。
3.2 Sigmoid
类型: SIGMOID
例子:
layers { name: "encode1neuron" bottom: "encode1" top: "encode1neuron" type: SIGMOID}
SIGMOID 层通过 sigmoid(x) 计算每一个输入x的输出,函数如下图。
3.3 TanH / Hyperbolic Tangent
类型: TANH
例子:
layers { name: "encode1neuron" bottom: "encode1" top: "encode1neuron" type: SIGMOID}
TANH层通过 tanh(x) 计算每一个输入x的输出,函数如下图。
3.3 Absolute Value
类型: ABSVAL
例子:
layers { name: "layer" bottom: "in" top: "out" type: ABSVAL}
ABSVAL层通过 abs(x) 计算每一个输入x的输出。
3.4 Power
类型: POWER
例子:
layers { name: "layer" bottom: "in" top: "out" type: POWER power_param { power: 1 scale: 1 shift: 0 }}
可选参数:
power [default 1]
scale [default 1]
shift [default 0]
POWER层通过 (shift + scale * x) ^ power计算每一个输入x的输出。
3.5 BNLL
类型: BNLL
例子:
layers { name: "layer" bottom: "in" top: "out" type: BNLL}
BNLL (binomial normal log likelihood) 层通过 log(1 + exp(x)) 计算每一个输入x的输出。
4. 数据层(Data Layers)
数据通过数据层进入Caffe,数据层在整个网络的底部。数据可以来自高效的数据库(LevelDB 或者 LMDB),直接来自内存。如果不追求高效性,可以以HDF5或者一般图像的格式从硬盘读取数据。
4.1 Database
类型:DATA
必须参数:
source:包含数据的目录名称
batch_size:一次处理的输入的数量
可选参数:
rand_skip:在开始的时候从输入中跳过这个数值,这在异步随机梯度下降(SGD)的时候非常有用
backend [default LEVELDB]: 选择使用 LEVELDB 或者 LMDB
4.2 In-Memory
类型: MEMORY_DATA
必需参数:
batch_size, channels, height, width: 指定从内存读取数据的大小
The memory data layer reads data directly from memory, without copying it. In order to use it, one must call MemoryDataLayer::Reset (from C++) or Net.set_input_arrays (from Python) in order to specify a source of contiguous data (as 4D row major array), which is read one batch-sized chunk at a time.
4.3 HDF5 Input
类型: HDF5_DATA
必要参数:
source:需要读取的文件名
batch_size:一次处理的输入的数量
4.4 HDF5 Output
类型: HDF5_OUTPUT
必要参数:
file_name: 输出的文件名
HDF5的作用和这节中的其他的层不一样,它是把输入的blobs写到硬盘
4.5 Images
类型: IMAGE_DATA
必要参数:
source:
text文件的名字,每一行给出一张图片的文件名和label
batch_size:
一个batch中图片的数量
可选参数:
rand_skip:
在开始的时候从输入中跳过这个数值,这在异步随机梯度下降(SGD)的时候非常有用
shuffle [default false]
new_height, new_width: 把所有的图像resize到这个大小
4.6 Windows
类型:WINDOW_DATA
4.7 Dummy
类型:DUMMY_DATA
Dummy 层用于development 和debugging。具体参数DummyDataParameter。
5. 一般层(Common Layers)
5.1 全连接层Inner Product
类型:INNER_PRODUCT
例子:
layers { name: "fc8" type: INNER_PRODUCT blobs_lr: 1 # learning rate multiplier for the filters blobs_lr: 2 # learning rate multiplier for the biases weight_decay: 1 # weight decay multiplier for the filters weight_decay: 0 # weight decay multiplier for the biases inner_product_param { num_output: 1000 weight_filler { type: "gaussian" std: 0.01 } bias_filler { type: "constant" value: 0 } } bottom: "fc7" top: "fc8"}
必要参数:
num_output (c_o):过滤器的个数
可选参数:
weight_filler [default type: 'constant' value: 0]:参数的初始化方法
bias_filler:偏置的初始化方法
bias_term [default true]:指定是否是否开启偏置项
通过全连接层后的大小变化:
输入:n * c_i * h_i * w_i
输出:n * c_o * 1 *1
5.2 Splitting
类型:SPLIT
Splitting层可以把一个输入blob分离成多个输出blobs。这个用在当需要把一个blob输入到多个输出层的时候。
5.3 Flattening
类型:FLATTEN
Flattening是把一个输入的大小为n * c * h * w变成一个简单的向量,其大小为 n * (c*h*w) * 1 * 1。
5.4 Concatenation
类型:CONCAT
例子:
layers { name: "concat" bottom: "in1" bottom: "in2" top: "out" type: CONCAT concat_param { concat_dim: 1 }}
可选参数:
concat_dim [default 1]:0代表链接num,1代表链接channels
通过全连接层后的大小变化:
输入:从1到K的每一个blob的大小n_i * c_i * h * w
输出:
如果concat_dim = 0: (n_1 + n_2 + ... + n_K) *c_1 * h * w,需要保证所有输入的c_i 相同。
如果concat_dim = 1: n_1 * (c_1 + c_2 + ... +c_K) * h * w,需要保证所有输入的n_i 相同。
通过Concatenation层,可以把多个的blobs链接成一个blob。
5.5 Slicing
The SLICE layer is a utility layer that slices an input layer to multiple output layers along a given dimension (currently num or channel only) with given slice indices.
5.6 Elementwise Operations
类型:ELTWISE
5.7 Argmax
类型:ARGMAX
5.8 Softmax
类型:SOFTMAX
5.9 Mean-Variance Normalization
类型:MVN