Neural networks are able to do with graphic images of previously unimaginable things. For example, in July 2018, NVIDIA demonstrated the algorithm that removes from images noise and superfluous labels. Unfortunately, ordinary users cannot take advantage of his skills, but there are developers who generously share their projects with ordinary people. One of them is a service that allows you to quickly remove the background from any photo with a person.
The service is called Remove.bg — it’s a great example of how machine learning-based projects become light consumer tools. Algorithms that quickly remove the background from photos that already exist and are available in the public domain. However, unlike them, this service provides the algorithm in the most convenient to use.
To remove the background image and leave only the main object, you just need to upload a file to a website or paste the direct link to the image — the algorithm itself will do the work and give a. PNG file with a transparent background. The tool only works with photos of people, so you can remove the background from pictures with animals and objects will not work.
At some points the service to cope with the removal of the even better Adobe Photoshop and similar programs. The difference is particularly noticeable on the photos of people with long hair, the neural network accurately identifies them and adds transparency.
Of course, the work encountered and inaccuracies. The algorithm does not always accurately defines the boundaries and is able to remove part of a person’s face. For example, the image below shows the remains of a dark background. However, the service is able to help in moments when you do not have photoshop or any other editor. The remaining defects can be corrected manually, and if you insert a PNG file into a different background, ugliness is can become invisible. .
To test the operation of the algorithm can be anyone on the website Remove.bg. The service is completely free.
If you liked the service? My opinion of him, you can write in the comments or in our Telegram chat.