This is the repo for Spatio-Temporal Garment Reconstruction Using Diffusion Mapping via Pattern Coordinates.
conda env create -f ./environment.yml
conda activate dmap
- Following the instruction to install Pymesh.
- Download model checkpoints from here, and put it under the root of the repo
./checkpoints.
- Download the data from here, and put it under the root of the repo
./data.
Use the scripts under ./scripts to recover 3D garment from the prepared images in ./data. The codes use the data from ./data as input, and save the results to ./fitting-results by default.
cd scripts
bash recover_all.sh
Step 1: synthesize the normal for the invisible back.
python ./step1_back_normal_guide.py
Step 2: predict the UV coordinates and depth from the normal.
python ./step2_uv_mapping_FB.py
Step 3: fit ISP to the incomplete mask to recover the complete mask/rest garment.
python ./step3_isp_seq_one_mask.py
Step 4: fit the diffusion prior to the incomplete UV positional maps to recover the complete UV maps/deformed garment.
python ./step4_uv_inpainting.py
Step 5: refine the recovered garment mesh using a set of observations and constraints (normal, mask, points, physical energy, temporal consistency, etc.).
python ./step5_post_opt.py
