An Iterative Approach for High-Quality Mask Generation in Image Matting

This is the official code of "An Iterative Approach for High-Quality Mask Generation in Image Matting (MPM)".

Prerequisites

• PyTorch 2.0
• Python 3.9
• You can use pip install requirements.txt to install the environment, but I wouldn't recommend it as I haven't tested it.

Datasets

You can put the data files in the MPM_MTM_Modules folder as well as in MPM_Mask_Acquire. In this article, we use AIM-500 and Distinctions-646 datasets for related experiments.

Usage

1. Generate Accurate Binary Mask
   • Open MPM_Mask_Acquire
   • Run the following command for iterative optimization from roughprompt to accurate mask.

   python scripts/PaintSeg.py --outdir $outdir$ --iters $iter_num$ --steps $diffusion step$ --dataset $dataset$ 
   

   • Put the obtained results in the MPM_MTM_Modules folder together with the datasets.
2. Use MPM_MTM_Modules
   • Run the main.py file for training.

    python main.py  --datasets {your datesets location} --fe {frozen encoder?} --norm {is norm your datasets image?}