Automatically Remove Backgrounds From Images

Researchers from Adobe, the Beckman Institute for Advanced Science and Technology and University of Illinois at Urbana-Champaign developed a deep learning-based method that clips objects from photos and videos.

Researchers have developed a number of different artificially intelligent programs to automatically subtract a background from an image, but most are based on colors. When presented with an image with similar colors such as the greens in a landscape photo, those programs tend to fail, Adobe says.

The AI method called “Deep Image Matting”, works by learning the structure of the “color” channel that contains all the transparencies in an image.

Using TITAN X GPUs and cuDNN with the Caffe deep learning framework, the researchers used nearly 50,000 images to train their network and 1,000 images to test the accuracy – the AI system is then able to guess the foreground and background elements of the training images and compare its guesses to the known reality of the images.

Although this is still under research, the AI method can possibly eliminate the need for green screen sets and eventually make its way inside one of Adobe’s photo-editing programs.

Read more >

12 thoughts on “Automatically Remove Backgrounds From Images

  1. Roderick on March 25, 2017 at 12:34 pm said:


  2. Imran Bughio on March 27, 2017 at 4:13 am said:

    Are they doing this research in closed doors or the code is open source for any one to get hands on?

    • David Dietrichstein on April 1, 2017 at 11:33 am said:

      Did you find out if they have released their code somewhere?

  3. marcioab on March 27, 2017 at 1:28 pm said:

    This is the #1 building block for visual AI. When this capability became available, an entire book can be generated from a single image. ( … and for autonomous cars, of course ).

  4. mpeniak on March 31, 2017 at 9:46 am said:

    Looks ok

  5. Antonio Kowatsch on March 31, 2017 at 4:49 pm said:

    This is nothing special. I’ve been doing this for years. My own project is written in C and achieves similar results.

    • Toke Faurby on April 1, 2017 at 6:53 am said:

      Pics, or it didn’t happen

    • David Dietrichstein on April 1, 2017 at 11:32 am said:

      Did you upload your code somewhere, maybe github? Could you share links to similar projects? I guess if this is not so hard to code, there should be a lot opensourced repos out there.

      • Roman Matěna on November 28, 2017 at 4:21 am said:

        i am also interested in similar projects

  6. Jeff Benshetler on April 1, 2017 at 11:34 am said:

    1) This method is achieves excellent results yet cannot run on a GPU on even modest sized images, e.g. 1k x 1k.
    2) What is the GPU run time performance for smaller images?

  7. Nate Gardner on May 1, 2017 at 1:37 pm said:

    Software like Vertus Fluid Mask and Topaz ReMask have been doing this for years, but cool!

    • sinan arik on October 16, 2017 at 6:35 am said:

      Nope. This says automatic. In such programs, users are generally needed to input what is going to be kept and what is going to be deleted.