When Janick Entremont, an Austrian photographer, decided to repost an image to instagram 300 times, this was the result:
And that’s just by the 80th iteration. While an extreme case, this experiment showcases the destructive nature of Instagram’s algorithm.
At Compression.ai, we decided to perform the same experiment with the same image, running through our machine learned algorithm 300 times. What we found, was that our algorithm converged extremely quickly. Within 4 iterations, it decided that it could not compress the image any longer without compromising its structural integrity. Any attempt beyond the 3rd iteration would simply return back the original image:
|
|
|
|
Original Image |
Compression.ai |
Compression statistics:
• Original image size: 192 KB
• Compressed image size: 73 KB
• Compression ratio: 62% (38% of the original size)
• MS-SSIM: 0.9886
• PSNR: 37.62
For more information on how we measure image quality at compression.ai and how this affects our algorithms, read the related blog post here.
