You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
2/ EfficientNets come to mind- I just have not been seeing the benefits relative to the classification performance. But this really applies generally, even within a set of architectures that have been trained the same way (FaceBook's wsl models, BiTM, etc).
3/ In those cases, you just don't really know until you actually try. You may be surprised, I can tell you that much! So don't just pick the one with the highest Top-1 accuracy in ImageNet...
Any better metrics that you might not have come up with along the way? Or sets of metrics?
1/ Just rules of thumb. So far I find the resnet architecture generally is better than EfficientNet. But the other thing that seems quite important is the training regime. The Big Transfer (BitM) paper spells it out pretty clearly that this makes a big difference in transfer ...
This. One of my most popular tweets was an offhand remark on how to eek out an extra 0.1% on ImageNet, which I regarded as useless b/c ‘interesting problems’ are not Imagenet, but the tweet *exploded* in popularity, which I found puzzling. Then I realized: they were all Kagglers
Oh well that makes perfect sense! Do you have a favorite architecture for your interesting work?
I don't know whether this is tested for CV, but worth a try.https://twitter.com/sai_prasanna/status/1265730581586866182?s=19 …
Sounds like a great lead: "Proper DNN comparison hence requires a comparison between their empirical score distributions on unseen data, rather than between single evaluation scores as is standard for more simple, convex models. "
Yeah, eeking out an extra bit of top-1 on ImageNet can be harmful to other tasks.... I don't have link handy, but there was a paper (neurips 2019 I think?) that showed training with label smoothing hurt downstream transfer learning ability...
If I correctly understood, you mean that we should not see any score(like accuracy). We should depend on the output result we are getting. Am I right?
Unfortunately that does seem to be the best way I know how to do it as of now! It's not like there's 0 correlation with accuracy but it's not great. I've spelled out rules of thumb that have seemed to be most useful in reply to this:https://twitter.com/MaxLenormand/status/1271156822976737280 …
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
