室内视频流高质量精确3DGS生成
1.Clone stellar_engine_web_reconstruction
git clone root@192.168.0.105:gs_scene/stellar_engine_web_reconstruction.git
cd stellar_engine_web_reconstruction
2.Create the environment
conda create -n stellar3d python=3.10.13 cmake=3.14.0 -y
conda activate stellar3d
conda install pytorch torchvision pytorch-cuda=11.8 -c pytorch -c nvidia # use the correct version of cuda for your system
pip install -r requirements.txt
pip install InstantSplat/submodules/simple-knn
pip install InstantSplat/submodules/diff-gaussian-rasterization
pip install InstantSplat/submodules/fused-ssim
pip install compressed_gaussians/submodules/diff-gaussian-rasterization
pip install compressed_gaussians/submodules/weighted_distance
3.Optional but highly suggested, compile the cuda kernels for RoPE (as in CroCo v2).
# DUST3R relies on RoPE positional embeddings for which you can compile some cuda kernels for faster runtime.
cd InstantSplat/croco/models/curope/
python setup.py build_ext --inplace
python stellar_web_recon_pipeline.py [-h] [-i INPUT] [-o OUTPUT] [-log LOG_DIR] [-downsample PCD_DOWNSAMPLE_RATIO] [-adc] [-iter GS_TRAIN_ITER]
[--save_iterations SAVE_ITERATIONS [SAVE_ITERATIONS ...]] [-v]
options:
-h, --help show this help message and exit
-i INPUT, --input INPUT
input video path,default /data/gs_scene/stellar_web_recon/videos/hcf_video3.mp4
-o OUTPUT, --output OUTPUT
output 3d model root dir,default /data/gs_scene/stellar_web_recon
-log LOG_DIR, --log_dir LOG_DIR
logfile directory
-downsample PCD_DOWNSAMPLE_RATIO, --pcd_downsample_ratio PCD_DOWNSAMPLE_RATIO
pointcloud downsample ratio, default 1
-adc, --densification
Adaptive Density Control (ADC)
-iter GS_TRAIN_ITER, --gs_train_iter GS_TRAIN_ITER
gaussian splatting train iterations，default 500
--save_iterations SAVE_ITERATIONS [SAVE_ITERATIONS ...]
-v, --verbose verbose