This repository provides a pipeline for fitting NeuS surface reconstruction from a Video input
Preprocess the captured video with COLMAP
python preprocess_data/preprocess_data.py --data $xxx.mp4 --output_dir $preprocess_dir
Rescale the scene
python preprocess_data/rescale_and_mask.py --work_dir $preprocess_dir
To generate object mask for the scene, refer to $U^2$-Net
Run the resonctruciton
CUDA_VISIBLE_DEVICES=1 python scripts/train_colmap.py \
--data_dir $preprocess_dir \
--config configs/default_colmap.yaml \
--train_dir ckpt/$xxx
--print_every 10 \
--export_mesh resultvis/$xxx/sdf_mc.ply
--ek_lambda 0.05
--n_epochs 3
--n_iters 5000
Get the mesh
CUDA_VISIBLE_DEVICES=1 python scripts/inference.py \
--data_dir $preprocess_dir \
--config configs/default_colmap.yaml \
--ckpt ckpt/$xxx/ckpt_3.pth