This week at GTC 2018 in San Jose, California, engineers from Adobe Applied Research and Tech (ART) and NVIDIA demonstrated a deep learning-based method that extracts the foreground content from its background without the use of a green screen.
During their demo, the team used a version of Photoshop with the “Deep Matte” algorithm.
The method called “Deep Image Matting”, uses deep learning to intelligently separate foreground from the background.
An example of the Deep Matte feature in Adobe Photoshop.
We first told you about this research last year. Since then, the team has moved to NVIDIA TITAN Volta GPUs and cuDNN based customized inferencing framework for optimal performance.
Though not publicly available, the team said “Deep Matte” could be released in a future version of Photoshop.
AI-Generated Summary
- Adobe Applied Research and Tech engineers demonstrated a deep learning method that extracts foreground content from backgrounds without a green screen at GTC 2018.
- The "Deep Matte" algorithm uses deep learning to intelligently separate foreground from background, based on the Deep Image Matting research paper.
- Since initial research disclosure, the team has moved to NVIDIA TITAN Volta GPUs and a cuDNN-based customized inferencing framework for optimal performance.
Next Steps
- Read the original research announcement for background on the technology.
- Explore NVIDIA deep learning resources to understand the underlying techniques.
- Review cuDNN documentation for the inferencing framework used in the demo.
Powered by NVIDIA Nemotron. AI-generated content may summarize information incompletely. Verify important information. Learn more
About the Authors
