diff options
| author | Matthew S <matthewsot@outlook.com> | 2017-01-29 23:50:11 -0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2017-01-29 23:50:11 -0800 |
| commit | ad37e620e15ce98abffc0d4edd41e5ddb5613360 (patch) | |
| tree | 783fa620ba8bfea90d9eb74dcfcdec47e4b559e8 | |
| parent | 966ac00fd7ad0b94e13fdb780e8c86966c9e4f6e (diff) | |
| -rw-r--r-- | README.md | 2 |
1 files changed, 2 insertions, 0 deletions
@@ -3,6 +3,8 @@ Extremely lossy image compression with neural networks Uses the limited, slow, and inefficient (but super simple and easy-to-use!) feed-forward neural network [Zoltar](https://github.com/matthewsot/zoltar). +*More important note:* There are much better ways of doing this, including with RNNs [as Google is doing](https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45534.pdf). This is a simple example using only a FFN. + *Note:* While this technique does seem to work rather well for specific photos and with specific configurations, it's mostly meant as a proof of concept. The image compression algorithms that power most JPEG compression will probably give you a better result for most applications and images. Feel free to check this out though! Maybe you can make it even better :) # Show me the numbers |