Skip to content

Commit 0f01d86

Browse files
committed
Update Chapter15
1 parent 0d1c978 commit 0f01d86

1 file changed

Lines changed: 13 additions & 2 deletions

File tree

15.Object_detection.ipynb

Lines changed: 13 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -222,6 +222,13 @@
222222
"R-CNN 的原作者 Ross Girshick 在吸收了 [SPPNet](https://arxiv.org/abs/1406.4729) 的优点后再一次推出了 R-CNN 的继承者 Fast R-CNN。Fast R-CNN 构思精巧,流程更为紧凑,大幅提升了目标检测的速度。与 R-CNN 相比,训练时间从 84 小时减少为 9.5 小时,测试时间从 47 秒减少为 0.32 秒,在 PASCAL VOC 2007 数据集上的准确率为 66%。"
223223
]
224224
},
225+
{
226+
"cell_type": "markdown",
227+
"metadata": {},
228+
"source": [
229+
"![img](images/chapter15/Fast_R-CNN.png)"
230+
]
231+
},
225232
{
226233
"cell_type": "markdown",
227234
"metadata": {},
@@ -242,18 +249,22 @@
242249
"\n",
243250
"- Softmax 层对 ROI 特征向量进行分类,回归层使用 Smooth L1 损失函数进行包围盒回归,并加权联合二者损失函数值\n",
244251
"\n",
252+
"$$ L_{loc}(t^u,v) = \\sum_{i \\in\\{x,y,w,h\\}} Smooth_{L1}(t_i^u-v_i) $$\n",
253+
"\n",
245254
"$$Smooth_{L1}(x)=\n",
246255
"\\begin{cases}\n",
247256
"0.5x^2&x < 1 \\\\\n",
248257
"|x|-0.5&otherwise\n",
249-
"\\end{cases}$$"
258+
"\\end{cases}$$\n",
259+
"\n",
260+
"$$ L(p,u,t^u,v) = L_{cls}(p,u)+\\lambda L_{loc}(t^u,v) $$"
250261
]
251262
},
252263
{
253264
"cell_type": "markdown",
254265
"metadata": {},
255266
"source": [
256-
"![img](images/chapter15/Fast_R-CNN.png)"
267+
"其中,$p$ 为预测分类概率,$u$ 为分类标签,$t^u=(t_x^u, t_y^u, t_w^u, t_h^u)$为预测为第 $u$ 类的包围盒,$v$ 为对应的包围盒标签。$L_{cls}(p,u)$ 为分类损失函数,$L_{loc}(t^u,v)$ 为包围盒损失函数,$L(p,u,t^u,v)$ 为总损失函数。"
257268
]
258269
},
259270
{

0 commit comments

Comments
 (0)