People who say deep learning has solved vision overlook video prediction, where the state of the art is still embarrassingly bad.
-
When training on ImageNet is still out of the question for many researchers' resources, you can't expect that much progress on video, where you could need a couple of orders of magnitude more resources.
You can easily train on ImageNet with a single cloud GPU, and you can do video prediction if you're smart about it. But the truth is we wouldn't know how to do it accurately even with infinite computing power.
-
-
You can train existing efficient architectures on 1 GPU. Not new research architectures.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.