GeoMaster is a state-of-the-art tool designed to enhance the geometric details of 3D models, providing high-resolution outputs suitable for various applications such as gaming, virtual reality, and 3D printing.
To get started with GeoMaster, follow these steps:
First, clone the GeoMaster repository to your local machine:
git clone https://github.com/hugoycj/GeoMaster.git --recursive
cd GeoMasterCreate the geomaster environment using conda:
conda create -n geomaster python=3.9
conda activate geomasterFirst, navigate to the submodule and install dependencies for submodules:
cd submodule/PromptDA
pip install -r requirements.txt
pip install -e .
cd ../nvdiffrast
pip install -e .Then, install the dependencies for this project (GeoMaster):
cd ../..
pip install -r requirements.txt
pip install -e .Then, Install Real-ESRGAN:
pip install git+https://github.com/sberbank-ai/Real-ESRGAN.gitAnd adjust the path: Geomaster/weights/RealESRGAN_x4.pth
Finally, install gaustudio from its GitHub repository:
pip install git+https://github.com/GAP-LAB-CUHK-SZ/gaustudioBefore running GeoMaster, ensure that your data is organized in the following structure:
.
├── confidence [Optional] Depth confidence maps
├── depths [Required] Depth maps
├── images [Required] RGB images
└── sparse [Required] SfM sparse reconstruction
First, you need to prepare the input data using the gm-prepare-data command:
gm-prepare-data -s ./examples/glt
After running gm-prepare-data, the following directory structure with essential files for subsequent mesh enhancing and other operations is generated:
.
├── confidence [Optional Input] Depth confidence maps
├── depths [Required Input] Raw depth maps
├── images [Required Input] RGB images
├── sparse [Required Input] SfM sparse reconstruction
│
├── mono_depths [Output] Monocular depth estimations
├── mono_depths_aligned [Output] Aligned monocular depths
├── normals [Output] Surface normal estimations
│
├── pc.ply [Output] Point cloud representation of the scene
└── mesh.ply [Output] Textured 3D mesh
└── trimesh.ply [Output] Untextured triangulated mesh
To enhance the geometric detail of your 3D model using gm-refine, use the following command:
gm-refine -s ${INPUT_DIR} -m ${MODEL_DIR}/${MODEL_NAME}.plyReplace:
${INPUT_DIR}with the path to your input data directory${MODEL_DIR}with the path to your model directory${MODEL_NAME}with the name of your model file
Example:
gm-refine -s examples/glt/ -m examples/glt.plyThe enhanced result will be saved as ${MODEL_DIR}/${MODEL_NAME}.refined.ply.
If you only have the mesh file and do not need to specify an input data directory, you can use the gm-process-mesh command to make the mesh watertight:
gm-process-mesh -m ${MODEL_DIR}/${MODEL_NAME}.plyReplace:
${MODEL_DIR}with the path to your model directory${MODEL_NAME}with the name of your model file
Example:
gm-process-mesh -m examples/glt.plyThe watertight result will be saved as ${MODEL_DIR}/${MODEL_NAME}.refined.ply, overwriting the previous refined mesh if it exists.
To remove invisible or rarely seen vertices from your model, use the following command:
gm-cull-mesh -s ${INPUT_DIR} -m ${MODEL_DIR}/${MODEL_NAME}.ply [--min_weights VALUE] [--use_mask]Replace:
${INPUT_DIR}with the path to your input data directory${MODEL_DIR}with the path to your model directory${MODEL_NAME}with the name of your model file
Optional parameters:
--min_weights VALUE: Set the minimum number of views a vertex must be visible in (default is 5)--use_mask: Apply masking during processing
Examples:
- Culling with minimum weights:
gm-cull-mesh -s examples/glt/ -m examples/glt.ply --min_weights 1- Culling with masking:
gm-cull-mesh -s examples/glt/ -m examples/glt.ply --use_mask- Culling with both minimum weights and masking:
gm-cull-mesh -s examples/glt/ -m examples/glt.ply --min_weights 5 --use_maskThe culled result will be saved as ${MODEL_DIR}/${MODEL_NAME}.clean.ply.
Feel free to contribute to GeoMaster by submitting issues or pull requests on our GitHub repository.
Happy Modeling with GeoMaster!