search

Res2Net / Res2Net-PretrainedModels

(ImageNet pretrained models) The official pytorch implemention of the TPAMI paper "Res2Net: A New Multi-scale Backbone Architecture"
https://mmcheng.net/res2net/
1.07k stars 215 forks source link

代码问题 #37

Closed swjtulinxi closed 4 years ago

swjtulinxi commented 4 years ago

self.nums = scale -1请问你论文是分的4组吧， 你代码为什么又分的三组呢

wangbin2018 commented 4 years ago

你看后面for i in range(self.nums): 所以0，1，2，3不就是四组

swjtulinxi commented 4 years ago

self.nums = scale -1，但是你前面是nums等于3啊，所以是0，1，2吧

width = int(math.floor(planes * (baseWidth/64.0))) 你这里假设planes=256，你的basewidth设置的26，那么width就等于104， spx = torch.split(out, self.width, 1) 那么就分成，104，104，48不是吗，这里也没有4份啊，求解释一下，我研究半天了，还是没弄明白啊

gasvn commented 4 years ago

Suppose there are 4 groups. Then three groups are hierarchically connected, and one group is linked to the output without processing. image