Semantic Image Matting

This is the official repository of Semantic Image Matting (CVPR2021).

Overview

Natural image matting separates the foreground from background in fractional occupancy which can be caused by highly transparent objects, complex foreground (e.g., net or tree), and/or objects containing very fine details (e.g., hairs). Although conventional matting formulation can be applied to all of the above cases, no previous work has attempted to reason the underlying causes of matting due to various foreground semantics.

We show how to obtain better alpha mattes by incorporating into our framework semantic classification of matting regions. Specifically, we consider and learn 20 classes of matting patterns, and propose to extend the conventional trimap to semantic trimap. The proposed semantic trimap can be obtained automatically through patch structure analysis within trimap regions. Meanwhile, we learn a multi-class discriminator to regularize the alpha prediction at semantic level, and content-sensitive weights to balance different regularization losses.

Dataset

Download our semantic image matting dataset (SIMD) here. SIMD is composed self-collected images and a subset of adobe images. To obtain the complete dataset, please contact Brian Price (bprice@adobe.com) for the Adobe Image Matting dataset first and follow the instructions within SIMD.zip.

Requirements

The codes are tested in the following environment:

Python 3.7
Pytorch 1.9.0
CUDA 10.2 & CuDNN 7.6.5

Performance

Some pretrained models are listed below with their performance.

Methods	SAD	MSE	Grad	Conn	Link
SIMD	27.9	4.7	11.6	20.8	model
Composition-1K (paper)	28.0	5.8	10.8	24.8
Composition-1K (repo)	27.7	5.6	10.7	24.4	model

Run

Download the model and put it under checkpoints/DIM or checkpoints/Adobe in the root directory. Download the classifier here and put it under checkpoints. Run the inference and evaluation by

python scripts/main.py -c config/CONFIG.yaml

Results

Reference

If you find our work useful in your research, please consider citing:

@inproceedings{sun2021sim,
  author    = {Yanan Sun and Chi-Keung Tang and Yu-Wing Tai}
  title     = {Semantic Image Matting},
  booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year      = {2021},
}

Acknowledgment

This repo borrows code from several repos, like GCA and FBA.

pgt4861/SIM