147 points by neural_compression 1 year ago flag hide 23 comments
johnsmith 1 year ago next
This is pretty cool! I've been experimenting with neural networks lately but I haven't thought about using them for image compression. I'd love to see some performance benchmarks.
nvidianinja 1 year ago next
I've benchmarked this neural network-powered image compression algorithm on my NVIDIA RTX 3080 and it outperforms all other current methods. Very promising technology.
linuxuser 1 year ago next
Nice to hear your success. Mind sharing your NVIDIA RTX 3080 benchmarks with this community?
rtx3080owner 1 year ago next
I'd be happy to share my results with the community. Just running a few more checks to make sure they are accurate.
hackerguy 1 year ago prev next
Definitely agree with the potential of using neural networks for image compression. Curious if anyone here has experience with other forms of machine learning in this space?
ml_expert 1 year ago next
Yes! I've worked with deep reinforcement learning for image compression and found it to be very effective. Highly recommend checking it out.
tensor_enthusiast 1 year ago prev next
Conveniently, TensorFlow offers a number of pre-trained models for image compression tasks, which could greatly simplify implementation.
iamnewatthis 1 year ago next
Is there any documentation on how to use TensorFlow for image compression?
learner 1 year ago next
Yes, the TensorFlow documentation is well-organized and easy to follow. I recommend starting with their tutorials to learn more.
darkmagician 1 year ago prev next
The TensorFlow site hasn't been as clear lately; I'd love a detailed guide on how to use their models for this particular implementation.
ml_guy 1 year ago next
I know TensorFlow can easily train models and load pre-trained models, but do they offer specific models for image compression?
sysadmin123 1 year ago prev next
Image compression will be vital for our company as we move towards storing pictures in the cloud. I'm looking forward to integrating this with our current infrastructure.
cpusucks 1 year ago prev next
For small images, this algorithm might have some overhead. Can't wait to see it evolve for handling slightly more complex data, like lower-res video compression.
programmergirl 1 year ago next
That's a good point -- neural network-powered image compression might too much overhead for real-time video applications.
visionary 1 year ago next
Implementing neural network-powered image compression in real-time video conferencing apps could really change the game.
coder 1 year ago prev next
Just wondering, how do the learned features compare with traditional image compression algorithms such as JPEG?
nn_nerd 1 year ago next
There's a great blog post on Medium comparing the learned features of nn-powered image compression vs. traditional methods. It's worth a read.
rnddev 1 year ago next
Agreed, the blog post was a great read. It showed me the potential of using neural networks for lossless compression.
compressionfan 1 year ago prev next
JPEG, PNG, and similar algorithms bring their trade-offs. The learned features of this nn-powered algorithm might be the path for better quality and performance in the future.
nn_friend 1 year ago next
True, better quality and performance is the goal all around. I'm excited to see how this will evolve in the long term!
xorjoint 1 year ago prev next
How the neural network-powered image compression handles varying levels of compression will be interesting so see.
student 1 year ago prev next
I wonder if it would be possible to train a neural network to compress already compressed images, thus increasing their compression ratio.
neuralnetworksftw 1 year ago prev next
I've read in a research paper that using reinforcement learning for neural compression can offer improvements in image quality and compression performance.