Skip to content

quick_start_en.md

Latest commit

 

History

History
120 lines (95 loc) · 3.94 KB

quick_start_en.md

File metadata and controls

120 lines (95 loc) · 3.94 KB
Raw

Quick Start

Installation

1. Install PaddlePaddle

Versions

  • PaddlePaddle >= 2.0.2

  • Python >= 3.7+

Due to the high computational cost of model, PaddleSeg is recommended for GPU version PaddlePaddle. CUDA 10.0 or later is recommended. See PaddlePaddle official website for the installation tutorial.

2. Download the PaddleSeg repository

git clone https://github.com/PaddlePaddle/PaddleSeg

3. Installation

cd PaddleSeg/Matting
pip install "paddleseg>=2.5"
pip install -r requirements.txt

Download pre-trained model

Download the pre-trained model in Models to pretrained_models. Take PP-MattingV2 as an example.

mkdir pretrained_models && cd pretrained_models
wget https://paddleseg.bj.bcebos.com/matting/models/ppmattingv2-stdc1-human_512.pdparams
cd ..

Prediction

export CUDA_VISIBLE_DEVICES=0
python tools/predict.py \
    --config configs/ppmattingv2/ppmattingv2-stdc1-human_512.yml \
    --model_path pretrained_models/ppmattingv2-stdc1-human_512.pdparams \
    --image_path demo/human.jpg \
    --save_dir ./output/results \
    --fg_estimate True

Prediction results are as follows:

Note: --config needs to match --model_path.

Background Replacement

export CUDA_VISIBLE_DEVICES=0
python tools/bg_replace.py \
    --config configs/ppmattingv2/ppmattingv2-stdc1-human_512.yml \
    --model_path pretrained_models/ppmattingv2-stdc1-human_512.pdparams \
    --image_path demo/human.jpg \
    --background 'g' \
    --save_dir ./output/results \
    --fg_estimate True

The background replacement effect is as follows:

Notes:

  • --image_path must be the specific path of an image.
  • --config needs to match --model_path.
  • --background can be passed into the background image path, or one of ('r','g','b','w'), representing a red, green, blue, or white background, default green if not passed.

Video Prediction

Run the following commad to predict the video, and remember to pass the video path by --video_path.

export CUDA_VISIBLE_DEVICES=0
python tools/predict_video.py \
    --config configs/ppmattingv2/ppmattingv2-stdc1-human_512.yml \
    --model_path pretrained_models/ppmattingv2-stdc1-human_512.pdparams \
    --video_path path/to/video \
    --save_dir ./output/results \
    --fg_estimate True

Prediction results are as follows:

Video Background Replacement

Run the following commad to replace video background, and remember to pass the video path by --video_path.

export CUDA_VISIBLE_DEVICES=0
python tools/bg_replace_video.py \
    --config configs/ppmattingv2/ppmattingv2-stdc1-human_512.yml \
    --model_path pretrained_models/ppmattingv2-stdc1-human_512.pdparams \
    --video_path path/to/video \
    --background 'g' \
    --save_dir ./output/results \
    --fg_estimate True

The background replacement effect is as follows:

Notes:

  • --background can be passed into the background image path, or background video path, or one of ('r','g','b','w'), representing a red, green, blue, or white background, default green if not passed.