Image Matting and Background Removal
Separating a foreground subject from its background at hair-level detail for photos and video.
Tier Grades
Tier A: Frontier AI labs
stretch
Mostly academic and small-company work.
Tier B: Indonesian AI labs
feasible
Training on public matting sets fits an Indonesian lab.
Tier C: Institution units such as ITB
feasible
A 7-author team produced BiRefNet; a university unit can match that.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Background removal for product photos is a quick retail proof of value.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataNot in the sources read.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 7 authors on the BiRefNet paper (arXiv 2401.03407).
- Time horizonYear.
- Shelf-life signalsPromptable segmenters reach matting quality, Image generators remove backgrounds as an edit
- Pipeline roleTier C and small firms publish matting models; Tier D uses them in e-commerce and media pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
BiRefNetvisionexemplar
RMBGvisionexemplar
Robust Video Mattingvisionexemplar
U2-NetvisionexemplarBEN (Background Erase Network)visionproduced by
GCA Mattingvisionproduced by
SINet (portrait segmentation)visionproduced by
IndexNetvisionproduced byDIM (Deep Image Matting)visionproduced byBASNetvisionproduced by
F3Netvisionproduced by
Universal matting (DAMO)visionproduced by
BodyPixvisionproduced byMediaPipe Selfie Segmentervisionproduced byBackground matting (OpenVINO public models)visionproduced byLinks
R&D TopicImage Matting and Background Removal
ModelsU2-NetUniversity of Alberta, visionBiRefNetNankai University, visionRMBGBria AI, visionRobust Video MattingUniversity of Washington, vision+11 more
U2-NetUniversity of Alberta, visionBiRefNetNankai University, visionRMBGBria AI, visionRobust Video MattingUniversity of Washington, vision+11 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 15
MediaPipe Selfie SegmenterGoogle, visionproduced byBodyPixGoogle, visionproduced byRMBGBria AI, visionexemplarRobust Video MattingUniversity of Washington, ByteDance, visionexemplarSINet (portrait segmentation)Naver, visionproduced byBiRefNetNankai University, visionexemplarF3NetInstitute of Computing Technology, Chinese Academy of Sciences, visionproduced byGCA MattingShanghai Jiao Tong University, visionproduced byIndexNetUniversity of Adelaide, visionproduced byU2-NetUniversity of Alberta, visionexemplarUniversal matting (DAMO)Alibaba DAMO, visionproduced byBackground matting (OpenVINO public models)visionproduced byPapers
- Deep Image MattingNing Xu et al., arXiv preprint, 2017 founding
- Bilateral Reference for High-Resolution Dichotomous Image SegmentationPeng Zheng et al., Bilateral Reference for High-Resolution Dichotomous Image Segmentation. CAAI Art, 2024 state of the art
- Robust High-Resolution Video Matting with Temporal GuidanceShanchuan Lin et al., arXiv preprint, 2021 state of the art
- U$^2$-Net: Going Deeper with Nested U-Structure for Salient Object DetectionXuebin Qin et al., arXiv preprint, 2020 state of the art
- MODNet: Real-Time Trimap-Free Portrait Matting via Objective DecompositionZhanghan Ke et al., arXiv preprint, 2020 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2401.03407 https://arxiv.org/abs/2401.03407
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Image Matting and Background Removal", https://aurion.semai.id/topics/image-matting-and-background-removal/, records of October 2026.