Instance-Segment-BASI

train

train 1

0. pool size = 50

1. pos_weight=5 训练segment 100000

2. pos_weight=3 训练segment 350000

3. add attention class 0.1 * loss_classes 100000

4. add attention class 0.2 * loss_classes 200000

5. 同时训练 attention class 和 segment 230000

6. result in class,segment,together

train 2

0. pool size = 90

1. pos_weight = 3 + attention class 0.2 * loss_classes 同时训练 attention class 和 segment 310000

2. result in begin/first

train 3

0. pool size = 90

1. segment 3 类

2. result in begin/second

train 4

0. segment border

1. segment 4 类

2. result in begin/third

train 5

0. coco

1. code in 5COCO

train 6

0. add attention

1. train by voc

GUI

BAISRunnerGUI.py

• 输入的三种方式：

  1. 图片路径
  2. 图片数据
  3. 网路路径

• 操作步骤

  1. click
  2. ok

• result

input	output

Idea

1. 边界和attention的权重

2. 用一个小的对抗网络判断是否完成

3. 对给定点所在的对象分割和分类、给两个点判断是否属于一个对象、给定点判断是否属于前景