diff options
author | Matthew S <matthewsot@outlook.com> | 2016-05-08 00:07:24 -0700 |
---|---|---|
committer | Matthew S <matthewsot@outlook.com> | 2016-05-08 00:07:24 -0700 |
commit | 966ac00fd7ad0b94e13fdb780e8c86966c9e4f6e (patch) | |
tree | 06fbb6766083d8acd7cd73e4963a1233a153b261 | |
parent | c6e9a1aa2ca82bab816a9b9a669179c78e5bd15c (diff) |
updated README to better explain the status of the project
-rw-r--r-- | README.md | 4 |
1 files changed, 3 insertions, 1 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). +*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 Original PNG image, **106,638 bytes** uncompressed and 106,795 bytes when compressed with the default Windows "send-to" compression: @@ -111,4 +113,4 @@ nimg reconstruct image.nimg To convert ``image.nimg`` back into a PNG with the 'missing' pixels highlighted in red (as ``image.reconstructed.png``): ``` nimg reconstruct image.nimg demo -```
\ No newline at end of file +``` |