LatComp (Latent Compression)
LatComp compression uses an AI model (VAE) and some custom code & math to compress images into a small, reversible format.
This work was inspired by Jeremy Howard and Jonathan Whitaker of fast.ai and answer.ai.
While taking the fast.ai course, I was learning about Variational Autoencoders (VAE) and began to wonder:
Is it possible to represent the latent space as an image, and then reconstruct the original image from that representation?
Compression Comparison: A 338 KB image compressed using various methods.
Size (KB) | 338 | 11 | 35 | 66 | 92 | 107 | 214 | 271 | 338 |
Note: Lossless compression means the original image can be perfectly reconstructed.
Use Cases:
- Save storage space
- Faster file transfers
- Backups & archives
Potential Improvements:
- Better/Faster AI model (VAE)
- Replace custom code & math with an AI model
- All-in-one AI Model
Compress your image into a small and reversible format.
Images bigger than 512x512 will be resized to reduce GPU memory usage.
Drop Image Here - or - Click to Upload
Examples
Get your original image back from a latcomp.
Drop Image Here - or - Click to Upload
Examples