Course project for Physically-Based Simulation in Computer Graphics HS21, ETH Zurich
Group 19: Yifan Yu, Bo Li, Yitian Ma
To setup the environment, use
pip install -r requirements.txt
You can change the simulation settings in src/simulator.py
global_params = {
'mode' : 'apic', # pic, apic, flip
'flip_weight' : 0.99, # FLIP * flip_weight + PIC * (1 - flip_weight)
'dt' : 0.01, # Time step
'g' : (0.0, 0.0, -9.8), # Body force
'rho': 1000.0, # Density of the fluid
'grid_size' : (64, 64, 64), # Grid size (integer)
'cell_extent': 0.1, # Extent of a single cell. grid_extent equals to the product of grid_size and cell_extent
'reconstruct_resolution': (100, 100, 100), # Mesh surface reconstruction grid resolution
'reconstruct_threshold' : 0.75, # Threshold of the metaball scalar fields
'reconstruct_radius' : 0.1, # Radius of the metaball
'num_jacobi_iter' : 100, # Number of iterations for pressure solving using jacobi solver
'damped_jacobi_weight' : 1.0, # Damping weighte in damped jacobi
'simulate_sand': False, # Simulate sand or water
'sand_dt': 0.001, # Time step for simulating sand
'scene_init': 0, # Choose from 0, 1, 2 to init the particle positions differently
}
The scene_init
parameter corresponds to: 0 - Centered dam break, 1 - Fall from sky, 2 - Cornered dam break.
Run with GGUI:
python main.py
During running with GUI, you can press Space
to reconstruct the mesh of current frame and export that into a obj model file. The output will reside in ../results
.
You can also run without GUI and reconstruct each frame:
python main.py 300
where the simulation contains 300 frames.
If the simulation dt is too small, add a second cmdline argument to reconstruct surface every few number of simulation steps:
python main.py 300 10
You can also export frames as PNGs to output/{method}_{scene_init}/
by changing OUTPUT
in src/main.py
to True
, the default frame number is 300. By default, each frame is acquired per 3 steps of fluid simulation, and 30 steps of sand simulation. The included GIFs in output-fluid
and output-sand
is generated by the output PNGs using FFMPEG.
- taichi==0.9.0
- numpy
- scipy
- PyMCubes
- connected-components-3d
PIC:
FLIP:
APIC:
Rendered:
water_low.mp4
PIC:
FLIP:
APIC:
Rendered: