1

I use Resnet-50 and ResNet-152 to implement my code, I read this:

We also note that ResNet-152 (3×+SK) is only marginally better than ResNet-152 (2×+SK),

So, I would like to undrestand the main difference between ResNet-152 (3×) and ResNet-152 (2×), meaning what corresponds 2 and 3 ? Thanks

Eliza
  • 584
  • 4
  • 14

2 Answers2

2

3x and 2x both mean "width". Larger width corresponds to higher number of parameters, meaning that more computations are performed.

According to this paper (Table 1), ResNet-152 (3×+SK) has 795 million parameters, while ResNet-152 (2×+SK) has only 354 million parameters (i.e. twice less). However, in terms of accuracy both perform almost equally.

If I were to choose between the two models, I would prefer the smaller one – ResNet-152 (2×+SK).

penkovsky
  • 893
  • 11
  • 14
0

When I find this :

Total params: 63,090,563 Trainable params: 4,719,619 Non-trainable params: 58,370,944

It's 2× or 3× ??

seni
  • 659
  • 1
  • 8
  • 20