def draw_goal_cornelbox(output, device=-1):
materials = {}
materials["light"] = DiffuseMaterial([1.0, 1.0, 1.0])
materials["white"] = DiffuseMaterial([0.5, 0.5, 0.5])
materials["green"] = DiffuseMaterial([0.0, 1.0, 0.0])
materials["red"] = DiffuseMaterial([1.0, 0.0, 0.0])
#shape_light = create_light(materials)
shape_floor = create_floor(materials)
shape_shortblock = create_shortblock(materials)
shape_tallblock = create_tallblock(materials)
shape = CompositeShape([shape_floor, shape_shortblock, shape_tallblock])
"""
light = np.array(GOAL_POS, dtype=np.float32)
model = ArrayLink(light)
light = PointLight(origin=model.data, color=[1, 1, 1])
"""
light = PointLight(origin=GOAL_POS, color=[1, 1, 1])
model = chainer.Link()
AP(model, 'data', light.origin)
fov = math.atan2(0.025, 0.035) * 180.0 / math.pi
camera = PerspectiveCamera(512, 512, fov, [278.0, 273.0, -800.0])
func = RaytraceFunc(shape=shape, light=light, camera=camera)
if device >= 0:
chainer.cuda.get_device_from_id(device).use()
model.to_gpu()
func.to_gpu()
y_data = func(1)
y_data = y_data.data
if device >= 0:
y_data = y_data.get()
cuda.get_device_from_id(device).synchronize()
img = y_data[0]
img = np.transpose(img, (1, 2, 0))
img = np.clip(img * 255, 0, 255).astype(np.uint8)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
cv2.imwrite(output, img)
return 0
def draw_start_cornelbox(output, device=-1):
materials = {}
materials["light"] = DiffuseMaterial([1.0, 1.0, 1.0])
materials["white"] = DiffuseMaterial([0.5, 0.5, 0.5])
materials["green"] = DiffuseMaterial([0.0, 1.0, 0.0])
materials["red"] = DiffuseMaterial([1.0, 0.0, 0.0])
shape_floor = create_floor(materials)
shape_shortblock = create_shortblock(materials)
shape_tallblock = create_tallblock(materials)
shape = CompositeShape([shape_floor, shape_shortblock, shape_tallblock])
fov = math.atan2(0.025, 0.035) * 180.0 / math.pi
camera = PerspectiveCamera(512,
512,
fov,
origin=[400.0, 300, -800.0],
direction=norm([-0.1, 0, 1]))
light = PointLight(origin=START_POS, color=[0.1, 0.1, 0.1])
func = RaytraceFunc(shape=shape, light=light, camera=camera)
if device >= 0:
chainer.cuda.get_device_from_id(device).use()
func.to_gpu()
y_data = func(1)
y_data = y_data.data
if device >= 0:
y_data = y_data.get()
cuda.get_device_from_id(device).synchronize()
img = y_data[0]
img = np.transpose(img, (1, 2, 0))
img = np.clip(img * 255, 0, 255).astype(np.uint8)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
cv2.imwrite(output, img)
return 0
def calc_goal_cornelbox(output, device=-1):
epoch = 50
outdir = os.path.dirname(output)
materials = {}
materials["light"] = DiffuseMaterial([1.0, 1.0, 1.0])
materials["white"] = DiffuseMaterial([0.5, 0.5, 0.5])
materials["green"] = DiffuseMaterial([0.0, 1.0, 0.0])
materials["red"] = DiffuseMaterial([1.0, 0.0, 0.0])
#shape_light = create_light(materials)
shape_floor = create_floor(materials)
shape_shortblock = create_shortblock(materials)
shape_tallblock = create_tallblock(materials)
shape = CompositeShape([shape_floor, shape_shortblock, shape_tallblock])
light = PointLight(origin=START_POS, color=[1, 1, 1])
model = chainer.Link()
AP(model, 'data', light.origin)
fov = math.atan2(0.025, 0.035) * 180.0 / math.pi
camera = PerspectiveCamera(512, 512, fov, [278.0, 273.0, -800.0])
func = RaytraceFunc(shape=shape, light=light, camera=camera)
if device >= 0:
chainer.cuda.get_device_from_id(device).use()
model.to_gpu()
func.to_gpu()
chainer.config.autotune = True
chainer.cudnn_fast_batch_normalization = True
#optimizer = optimizers.Adam(alpha=1e-1, beta1=0.9, beta2=0.999, eps=1e-08)
optimizer = optimizers.SGD(lr=0.001)
optimizer.setup(model)
#dataset
train_dataset = RaytraceDataset(output)
train_iter = chainer.iterators.SerialIterator(train_dataset,
1,
shuffle=True)
#updator
updater = RaytraceUpdater(train_iter,
model,
func,
optimizer,
outdir,
device=device)
#trainer
trainer = training.Trainer(updater, (epoch, 'epoch'), outdir)
log_interval = (1, 'iteration')
trainer.extend(extensions.dump_graph('main/loss'))
trainer.extend(extensions.LogReport(trigger=log_interval))
trainer.extend(extensions.PrintReport([
'epoch',
'iteration',
'main/loss',
'pos/light_x',
'pos/light_y',
'pos/light_z',
]),
trigger=log_interval)
trainer.extend(extensions.ProgressBar(update_interval=1))
trainer.run()
return 0
def calc_goal_cornelbox(output, device=-1):
epoch = 100
outdir = os.path.dirname(output)
materials = {}
materials["light"] = DiffuseMaterial([1.0, 1.0, 1.0])
materials["white"] = DiffuseMaterial([0.5, 0.5, 0.5])
materials["green"] = DiffuseMaterial([0.0, 1.0, 0.0])
materials["red"] = DiffuseMaterial([1.0, 0.01, 0.01])
shape_floor = create_floor(materials)
shape_shortblock = create_shortblock(materials)
shape_tallblock = create_tallblock(materials)
shape = CompositeShape([shape_floor, shape_shortblock, shape_tallblock])
# origin=[278.0, 273.0, -800.0], direction=[0, 0, 1])
# origin=[400.0, 300, -800.0], direction=norm([-0.1, 0, 1])
fov = math.atan2(0.025, 0.035) * 180.0 / math.pi
camera = PerspectiveCamera(512,
512,
fov,
origin=[270.0, 273.0, -800.0],
direction=norm([-0.1, 0, 0.8]))
light = PointLight(origin=START_POS, color=[0.1, 0.1, 0.1])
func = RaytraceFunc(shape=shape, light=light, camera=camera)
model1 = chainer.Link()
AP(model1, 'camera_position', camera.origin)
model2 = chainer.Link()
AP(model2, 'camera_zaxis', camera.zaxis)
AP(model2, 'camera_xaxis', camera.xaxis)
AP(model2, 'camera_yaxis', camera.yaxis)
if device >= 0:
chainer.cuda.get_device_from_id(device).use()
model1.to_gpu()
model2.to_gpu()
func.to_gpu()
chainer.config.autotune = True
chainer.cudnn_fast_batch_normalization = True
optimizer1 = O.SGD(lr=1e-3)
optimizer1.setup(model1)
optimizer2 = O.SGD(lr=1e-7)
optimizer2.setup(model2)
#dataset
train_dataset = RaytraceDataset(output)
train_iter = chainer.iterators.SerialIterator(train_dataset,
1,
shuffle=True)
dict_models = {'position': model1, 'direction': model2}
dict_optimizers = {'position': optimizer1, 'direction': optimizer2}
#updator
updater = RaytraceUpdater(train_iter,
dict_models,
func,
dict_optimizers,
outdir,
device=device)
#trainer
trainer = training.Trainer(updater, (epoch, 'epoch'), outdir)
log_interval = (1, 'iteration')
trainer.extend(extensions.dump_graph('main/loss'))
trainer.extend(extensions.LogReport(trigger=log_interval))
trainer.extend(extensions.PrintReport([
'epoch', 'iteration', 'main/loss', 'camera_position/x',
'camera_position/y', 'camera_position/z', 'camera_direction/x',
'camera_direction/y', 'camera_direction/z'
]),
trigger=log_interval)
trainer.extend(extensions.ProgressBar(update_interval=1))
trainer.run()
return 0