search

megvii-model / SinglePathOneShot

MIT License
259 stars 53 forks source link

Evolution search results #6

Open ultmaster opened 4 years ago

ultmaster commented 4 years ago

Hi. Sorry to bother again.

I tried to run your evolution search code, with checkpoints-150000 provided. The results look like this:

image

which is nowhere near the results shown in paper (over 68 in general).

image

Trying to dig out why, I found that recalculating BN on 20k images brings little improvement (less than one percent) to validation performance, compared to inheriting BN directly from the supernet. I'm not sure whether there is something wrong here. Have you guys observed a similar phenomenon?

Or, my wild guess is, the supernet you have shared is not a well-trained supernet?

haiduo commented 6 months ago

我感觉不用BN重新标定，搜索的子网的精度反而会更好，也许作者为了筛选出更好的，所以需要BN标定？