stride = 16 or (4, 8, 16, 32, 64) ? and why?

[dongzhenguo2016]

在cascade_r101v1_fpn_1x.py中的

class RpnParam:
class anchor_generate:
scale = (8,)
ratio = (0.5, 1.0, 2.0)
stride = (4, 8, 16, 32, 64) # 此处为(4, 8, 16, 32, 64)
image_anchor = 256
max_side = 1400

但是在tridentnet_r101v1c4_c5_1x.py中的

class RpnParam:
class anchor_generate:
scale = (2, 4, 8, 16, 32)
ratio = (0.5, 1.0, 2.0)
stride = 16 # 此处为16
image_anchor = 256
为什么有这种区别呢？为什么在tridentnet中stride 不能等于 (4, 8, 16, 32, 64)呢？
这个stride是什么作用呢？
thanks everyone!
happy 2020!

[huangzehao]

In FPN, we generate anchors on multiple feature maps which have different strides.
In tridentnet, we only generate anchors on a single feature map whose stride is 16.

[dongzhenguo2016]

In FPN, we generate anchors on multiple feature maps which have different strides.
In tridentnet, we only generate anchors on a single feature map whose stride is 16.
thanks very much!