Skip to content

Latest commit

 

History

History
38 lines (28 loc) · 2.15 KB

README_rgb_fitting.md

File metadata and controls

38 lines (28 loc) · 2.15 KB
Raw

RGB Fitting

RGB fitting from a single face image

  • Put the input images into a folder.
  • Modify the configuration and then run the following script for fitting.
run_rgb_fitting.sh  # Please refer to this script for detailed configuration

Reproducing REALY benchmark results

  • For the REALY benchmark results, all the fitting-based methods compared in the paper (Table 5), used 86 landmarks during the fitting process.
  • We provide the preprocessed data of REALY benchmark, which contains the detected 86 landmarks. The file (named "fitting_realy.zip") can be downloaded from:
  • Modify the configuration and then run the following script for fitting REALY images.
run_rgb_fitting_realy.sh  # Please refer to this script for detailed configuration
  • To calculate the evaluation metrics of REALY benchmark, please follow the instructions of the official repository. One should sign Agreement for downloading the benchmark data.

A new version of RGB fitting

  • The changes of RGB fitting process
    • Instead of using FFHQ-UV dataset, we use the FFHQ-UV-Interpolate dataset to train the GAN-based texture decoder.
    • When training the GAN-based texture decoder, instead of generating the entire UV-texture map (1024x1024), this version only generates the texture of the facial area (cut out the 630x630 facial area, and then resize to 1024x1024).
    • During RGB fitting process, the output texture map is first blended with template UV-texture map, and then resized back to original resolution.
  • Modify the configuration and then run the following script for fitting.
run_rgb_fitting_cropface630resize1024.sh  # Please refer to this script for detailed configuration
  • In some samples, this version is able to generate UV-texture maps with more details, such as the sample below.

fitting samples