YOLO
1. key word:
one pass solves all:
-
divide image into several grids
-
for each grid: predict bounding boxes & a class probabilities vector at the same time
-
each bounding box:
(x, y) : 相较于grid cell中心的offset
(w, h) : bounding box的长和宽
C : confidence score, 此bounding box包含物体的概率 -
P ( c i ∥ O b j e c t ) P(c_i\|Object) P(ci∥Object) :在此grid cell所属的bounding boxes包含物体的情况下,这个物体属于哪一类别
-
-
例子:输入图片分为 7 ∗ 7 7*7 7∗7的网格,每一格预测两个bounding boxes,物体一共有10类:
输出的维度 = 7 ∗ 7 ∗ ( 2 ∗ 5 + 10 ) 7*7*(2*5+10) 7∗7∗(2∗5+10)