import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import make_axes_locatable
import lmfunctest as ft
import lmscene
import lightmetrica as lm
os.getpid()
# %load_ext lightmetrica_jupyter
lm.init('user::default', {})
lm.parallel.init('parallel::openmp', {'numThreads': -1})
lm.log.init('logger::jupyter', {})
lm.info()
scenes = lmscene.scenes_small()
rmse_series = pd.Series(index=scenes)
for scene in scenes:
print("Testing [scene='{}']".format(scene))
lm.reset()
lm.asset('film_output', 'film::bitmap', {'w': 1920, 'h': 1080})
# Load scene and render
lmscene.load(ft.env.scene_path, scene)
lm.build('accel::sahbvh', {})
lm.render('renderer::raycast', {'output': lm.asset('film_output')})
img_orig = np.copy(lm.buffer(lm.asset('film_output')))
accel = lm.load_accel('accel', accel_name)
scene.set_accel(accel.loc())
scene.build()
film = lm.load_film('film_output', 'bitmap', w=1920, h=1080)
renderer = lm.load_renderer('renderer',
'raycast',
scene=scene,
output=film)
renderer.render()
return np.copy(film.buffer())
# + {"code_folding": []}
# Accels and scenes
accel_names = ['nanort', 'embree', 'embreeinstanced']
scene_names = lmscene.scenes_small()
# +
def rmse(img1, img2):
return np.sqrt(np.mean((img1 - img2)**2))
def rmse_pixelwised(img1, img2):
return np.sqrt(np.sum((img1 - img2)**2, axis=2) / 3)
# -
# Execute rendering for each scene and accel
rmse_df = pd.DataFrame(columns=accel_names, index=scene_names)
