654 points by ai_researcher 5 months ago flag hide 9 comments
johndoe 5 months ago next
This is quite interesting! I wonder how it differs from current architectures.
quantumhipster 5 months ago next
From my understanding, the authors mentioned they used open-source libraries, so there's a good chance it'll be usable with those frameworks.
aienthusiast 5 months ago prev next
I've looked into the paper, and it seems like the authors have made some significant improvements. The ablation studies are on point.
dataninja 5 months ago next
Could you elaborate more on the improvements? I'm curious how it might impact real-world applications.
techjunkie 5 months ago prev next
Anyone knows if this is compatible with TensorFlow or PyTorch? Would love to test it out for myself.
janedoe 5 months ago next
I know TensorFlow has good support for custom models and layers, so this should work great with it.
haskellprodigy 5 months ago next
From the paper, they seemed to focus more on a Theano implementation, but TensorFlow and PyTorch support should appear soon with more community engagement.
opensourcelover 5 months ago prev next
This makes me really excited for the future of image recognition, and I hope the authors open-source their code.
pseudorand 5 months ago next
Given the nature of their research and community involvement, I think the chances of open-source code are high. Just a matter of time, I guess!