A Trimap-Free Portrait Matting Solution in Real Time [AAAI 2022]

Go to file

actboy a070aaeeed full train demo		2022-02-13 16:14:29 +08:00
demo	(#161 ) - Make the file `demo/image_matting/colab/inference.py` can run on CPU	2022-01-11 00:53:34 +08:00
doc/gif	update GIF	2021-06-20 04:57:05 +08:00
onnx	add ONNX version of MODNet	2021-02-19 14:48:22 +08:00
pretrained	预训练模型	2022-02-10 19:31:50 +08:00
src	full train demo	2022-02-13 16:14:29 +08:00
torchscript	support TorchScript	2021-03-12 17:10:47 +08:00
.gitignore	调试模型训练代码	2022-02-09 21:45:36 +08:00
LICENSE	Create LICENSE	2022-01-28 14:37:41 +08:00
README.md	full train demo	2022-02-13 16:14:29 +08:00

README.md

说明

代码fork from MODNet官方代码。本项目完善了数据准备、模型评价及模型训练相关代码

模型训练、评价、推理

# 1. 下载代码并进入工作目录
git clone https://github.com/actboy/MODNet
cd MODNet

# 2. 安装依赖
pip install -r src/requirements.txt

# 3. 下载并解压数据集
wget -c https://paddleseg.bj.bcebos.com/matting/datasets/PPM-100.zip -O src/datasets/PPM-100.zip
unzip src/datasets/PPM-100.zip -d src/datasets

# 4. 训练模型
python src/trainer.py

# 5. 模型评估
python src/eval.py

# 6. 模型推理
python src/infer.py