Skip to content

Latest commit

 

History

History
97 lines (85 loc) · 3.99 KB

README_ckp_topo.md

File metadata and controls

97 lines (85 loc) · 3.99 KB

Download

The checkpoints used in FFHQ-UV project

File structure

|--FFHQ-UV  
    |--checkpoints 
        |--arcface_model
            |--ms1mv3_arcface_r50_fp16_backbone.pth
        |--deep3d_model
            |--epoch_latest.pth
        |--dlib_model
            |--shape_predictor_68_face_landmarks.dat
        |--dpr_model
            |--trained_model_03.t7
        |--e4e_model
            |--e4e_ffhq_encode.pt
        |--exprecog_model
            |--FacialExpRecognition_model.t7
        |--lm_model
            |--68lm_detector.pb
        |--mtcnn_model
            |--mtcnn_model.pb
        |--parsing_model
            |--79999_iter.pth
        |--resnet_model
            |--resnet18-5c106cde.pth
        |--styleflow_model
            |--modellarge10k.pt
            |--expression_direction.pt
        |--stylegan_model
            |--stylegan2-ffhq-config-f.pkl
        |--texgan_model
            |--texgan_ffhq_uv.pth
            |--texgan_cropface630resize1024_ffhq_uv_interpolate.pth
        |--vgg_model
            |--vgg16.pt

The introduction of these checkpoints

  • arcface_model: Arcface backbone, download from Arcface.
  • deep3d_model: Our trained Deep3D model with shape basis HiFi3D++.
  • dlib_model: The 68 landmarks detector from dlib.
  • dpr_model: The checkpoint of DPR.
  • e4e_model: The checkpoint of encoder4editing.
  • exprecog_model: The face expression recognition from this repo.
  • lm_model: The 68 landmarks detector from Deep3D.
  • mtcnn_model: The 5 point detector from MTCNN.
  • parsing_model: The face parsing from this repo.
  • resnet_model: ResNet backbone, download from this link.
  • styleflow_model: The checkpoint of StyleFlow and the expression editing direction found by SVM.
  • stylegan_model: The checkpoint of StyleGAN2.
  • texgan_model: Our Texture GAN models trained on FFHQ-UV dataset or FFHQ-UV-Interpolate dataset.
  • vgg_model: VGG backbone, download from this link.

The topology assets used in FFHQ-UV project

File structure

|--FFHQ-UV  
    |--topo_assets 
        |--center_face_mask.png
        |--hair_mask.png
        |--major_valid_front_mask.png
        |--major_valid_left_mask.png
        |--major_valid_right_mask.png
        |--major_valid_whole_mask.png
        |--minor_valid_front_mask.png
        |--minor_valid_left_mask.png
        |--minor_valid_right_mask.png
        |--minor_valid_whole_mask.png
        |--mouth_constract_mask.png
        |--nosal_base_mask.png
        |--nostril_mask.png
        |--template_base_uv.png
        |--unwrap_1024_info_mask.png
        |--hifi3dpp_model_info.mat
        |--hifi3dpp_mean_face.obj
        |--similarity_Lm3D_all.mat
        |--unwrap_1024_info.mat

The introduction of these topology assets

  • *_mask.png: The masks used in texture UV-map unwrapping.
  • hifi3dpp_model_info.mat: The HiFi3D++ face model information.
  • hifi3dpp_mean_face.obj: The mean face of the HiFi3D++ 3DMM shape basis.
  • similarity_Lm3D_all.mat: The 68 3D landmarks on the mesh.
  • unwrap_1024_info.mat: The coordinates used in texture UV-map unwrapping.