I really don’t like compressed photos or videos. It is one of the major drawbacks of social media networks and it is understandable why they do this. Although these photos and videos look way better when uncompressed, compressing media saves space on the servers which is a much needed cost saving measure for these companies.
Twitter has over 310 million active users who upload photos regularly, so the company has to compress the photos to save space. A perfect situation for the company is the ability to compress what its users are uploading to their servers, but also the photos should not bear that obvious disadvantage of compression where they significantly lose their quality.
That is why one of its acquisitions was able to come up with a machine learning system that was able to compress the photos that looked better than what the traditional JPEG compression looked like. The acquisition in question is Magic Pony which Twitter acquired in June, apparently was able to create an algorithm that compresses photos more efficiently than JPEG2000 (newer version of JPEG).
The algorithm works by looking at high quality images and creates rules for recreating an image with less information. Since is based on machine learning, it actually comes up with its own methods by checking patterns to represent the data more effectively. However, there is the issue of how we as humans perceive as “high quality” and how computers view the same variable, so the Twitter team decided to go with how humans perceive this.
Apparently, it was a success. A group of people were shown pictures of successful and poorly compressed pictures and later own shown 273 images that were compressed by the three algorithms (JPEG2000, JPEG and Twitter’s). On average, it was reported that Twitter’s algorithm was rated better than the other formats.
What does this mean? Well it means that if Twitter implements this to their assets, we will be able to upload photos which after undergoing compression, will look much better than before.