150 points by alex_coder 4 months ago flag hide 9 comments
username2 4 months ago next
I wonder how the power consumption of these algorithms compares to current solutions...
username1 4 months ago prev next
This is impressive work! Real-time image recognition has many exciting applications.
username1 4 months ago next
@username2 Good question! The team mentioned in the paper that power efficiency was, in fact, one of their key considerations.
username3 4 months ago prev next
So, how does the accuracy compare to XYZ-algo? I realize that's comparing apples to oranges, but I'm curious.
username2 4 months ago next
@username3 From the testing mentioned in the article, the new algorithms come out slightly ahead in terms of accuracy. However a direct comparison is tough since the edge cases used for XYZ algorithm may not apply here.
username4 4 months ago prev next
Has anyone built any practical applications on top of this? Can't wait to see real-world examples!
username5 4 months ago next
@username4 Some researchers mentioned possible integration with autonomous vehicle guidance systems and medical imaging tools. Exciting times are ahead!
username6 4 months ago prev next
What tooling is required to implement it? Does it support Tensorflow/Keras/Pytorch? Can't wait to try this out myself.
username1 4 months ago prev next
The research paper mentions that porting the implementation to popular deep learning frameworks should be straightforward, thanks to a modular design.