Computer Vision & Graphics Machine Learning & Data Science Popular

Adobe’s DL-Based ‘HDMatt’ Handles Image Details Thinner Than Hair

UIUC, Adobe Research and University of Oregon propose HDMatt, a Deep Learning-based image matting Cross-Patch Context module for high-resolution image inputs.

Image matting plays a key role in image and video editing and composition. Although existing deep learning approaches can produce acceptable image matting results, their performance suffers in real-world applications, where the input images are mostly high resolution. To address this, a group of researchers from UIUC, Adobe Research and the University of Oregon have proposed HDMatt, the first deep learning-based image matting approach for high-resolution image inputs.

Generally, deep learning approaches take an entire input image and an associated trimap to infer the alpha matte using convolutional neural networks. Such methods however may fail when dealing with high-resolution input images in sizes of 5000×5000 pixels or higher due to hardware limitations.


The researchers designed HDMatt to crop an input image and trimap into patches, then estimate the alpha values of each patch. Considering the information loss while only using a single patch and the prediction inconsistency between different patches, HDMatt introduces a novel Cross-Patch Context module (CPC) to effectively leverage cross-patch information for each query (current) patch. The estimated alpha values of each patch are then stitched together to output the final alpha matte of the entire image.

Cross-Patch Context (CPC) module workflow comprises a context patch sampling and a Trimap-guided Non-Local (TGNL) operation.

The team tested HDMatt’s capability using the Adobe Image Matting (AIM) and AlphaMatting benchmarks, where its quantitative results were all superior to existing SOTA approaches.


The team also conducted comparative evaluations with SOTA image matting methods IndexNet and ContexNet using input images with resolutions of up to 6000×6000 pixels, in which HDMatt was able to extract finer and more accurate details.

Visual comparison on real-world HR images. Image sizes from top to bottom: 5616 × 3744, 5779 × 3594, 4724 × 3929.

The paper High-Resolution Deep Image Matting is on Arxiv. Notably, second author Ning Xu from Adobe Research was first author on the 2017 paper Deep Image Matting.

Analyst: Victor Lu | Editor: Michael Sarazen; Fangyu Cai


Synced Report | A Survey of China’s Artificial Intelligence Solutions in Response to the COVID-19 Pandemic — 87 Case Studies from 700+ AI Vendors

This report offers a look at how China has leveraged artificial intelligence technologies in the battle against COVID-19. It is also available on Amazon KindleAlong with this report, we also introduced a database covering additional 1428 artificial intelligence solutions from 12 pandemic scenarios.

Click here to find more reports from us.

AI Weekly.png

We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Synced Global AI Weekly to get weekly AI updates.

6 comments on “Adobe’s DL-Based ‘HDMatt’ Handles Image Details Thinner Than Hair

  1. Pingback: Adobe’s DL-Based ‘HDMatt’ Handles Image Details Thinner Than Hair - GistTree

  2. I figured it was a fantasy that I was not able to pursue anywhere; now I believe that Adobe is fascinated by valuable findings. Give us the latest possibilities; please go ahead.

  3. Pingback: Adobe Research Proposes HDMatt, A Deep Learning-Based Image Matting Approach | MarkTechPost

  4. Pingback: Adobe Research Proposes HDMatt, A Deep Learning-Based Image Matting Approach – The Best

  5. Pingback: Adobe Research Proposes HDMatt, A Deep Learning-Based Image Matting Approach – Best Trendin'

  6. Really informative and helpful tips! These tips have been of great benefit to my work. I would like to get something more modern from this site. Thanks For Sharing and Keep Up the Good Work.

Leave a Reply

Your email address will not be published. Required fields are marked *

%d bloggers like this: