Skip to content

xianfenggao/zgc-aisc2022

Repository files navigation

AISC 2022 预赛code

Method

主要方法来自于论文 https://arxiv.org/abs/2106.15058, 外加perceptual loss 以及feature loss

生成模型使用TFace自己训练的模型: TFace_IR101_Aug_glint360k,MXNET_LResNet100E_IR_SGM,TFace_DenseNet201_splitbn_faceemore_glintasia,TFace_MobileFaceNet_Aug_faceemore,TFace_MobileFaceNet_Aug_vggface2,TFace_MobileFaceNet_Aug_umd,TFace_MobileFaceNet_Aug_webface

训练数据集来自于InsightFace, 训练代码来自于TFace(https://github.com/Tencent/TFace)

mask

测试了很多种方法,最终选择了自己手绘的mask,如下: image

测试脚本

run_generate.sh

About

code for aisc 2022 competition

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages