- Windows / Ubuntu (WSL tested only)
- Anaconda
- Build Skinweights Project in
Canonicalization/
. - Build AffectWeights Project in
Canonicalization/
. - Build MatchSMPL Project in
Canonicalization/FitSMPL_python/
. - Build deepanim Project in
data_process/
. - Set up the conda environment by running the command
conda env create -f AITS.yml
in theenv/
folder. Then , activate the environment withconda activate AITS
.
- Place the folder
datasets_template/data_name
in an appropriate location and renamedata_name
to your desired name. - Place input scans inside
data_name/data/src_meshes
folder (it is recommended to name them "0001.obj", "0002.obj", etc., but the extension can be.obj
or.ply
). - Execute
python .\Pose2Texture\utils\resample.py --path "path\to\[data_name]"
. If successful, the resampled meshes will be generated and placed in thedata\resampled_meshes\
folder. - Follow the instructions in FitSMPL_python_README to initialize SMPL fitting.
- Change the variable
Dataset_path
indata_make.bat
to the path ofdata_name
, and input the dataset size insize
. Then, Rundata_make.bat
.
If you get an error on the way, comment it out with@REM
or something similar and re-run from the point where it stopped.
To perform training, testing, and evaluation, execute a bat file in the Pose2Texture
folder:
- pop_itp.bat: Interpolation testing using the dataset provided by pop
- pop_etp.bat: Extrapolation testing using the dataset provided by pop
- snarf_itp.bat: Interpolation testing using the dataset provided by snarf
- snarf_etp.bat: Extrapolation testing using the dataset provided by snarf
The bat file performs the following steps:
- Creation of displacement texture
- Network training
- Testing
- Evaluation
In addition, it is necessary to add and edit the yaml file with the conf name specified in the bat file (e.g., Cape_pop_blazerlong_volleyball_itp
for pop_itp.bat
) in Pose2Texture/conf/train
.
You can check the training loss graphs etc. by running mlflow ui
in Pose2Texture/
. (mlflow)
Our own 4D dataset is available for download with fitted SMPL parameters at the following url: https://archive.iii.kyushu-u.ac.jp/public/FjIrAHGJbYBLDlpRzQ_R5gt2Xg8nMQ0F-_0wxgka-jzD
(内部者へ) Gitlabに付随のWikiに詳細を記載しています.公開する際には、公開すべきでない情報(パス等)が含まれていないかの確認をお願いします。