class PaddedColorToImage(Node):
"""
Takes a color input and creates a 4x4 texture from it, where one corner is assigned
a color, while the rest is black.
Inputs:
color:
A color tuple input.
Outputs:
image:
A valid picture of the size 4x4, which can be used as a texture or for
something else.
"""
color = NodeInput((1, 1, 1), ntype=T.RGB())
image = NodeOutput(ntype=T.Image())
def update(self):
image = (
(torch.tensor(self.color) * 2 - 1)
.unsqueeze(0)
.unsqueeze(2)
.unsqueeze(2)
.float()
.to(self.device)
)
self.image = f.pad(image, (1, 0, 1, 0), value=-1)
class ColorToImage(Node):
"""
Takes a color input and creates a valid texture from it.
Inputs:
color:
A color tuple input.
Outputs:
image:
A valid picture of the size 1x1, which can be used as a texture or for
something else.
"""
color = NodeInput((1, 1, 1), ntype=T.RGB())
image = NodeOutput(ntype=T.Image())
def update(self):
image = (
(torch.tensor(self.color) * 2 - 1)
.unsqueeze(0)
.unsqueeze(2)
.unsqueeze(2)
.float()
.to(self.device)
)
self.image = image
class ModelMatrix(Node):
"""
Generates a translation matrix from all possible input variables.
Inputs:
position:
A tuple of (x,y,z) position coordinates.
rotation:
A tuple of (x,y,z) rotation data. Not safe against gimbal lock!
scale:
A tuple of(x,y,z) scaling values.
Outputs:
model_matrix:
A translation matrix output.
"""
position = NodeInput((0.0, 0.0, 0.0), ntype=T.Vector3(-5.0, 5.0))
rotation = NodeInput((0.0, 0.0, 0.0), ntype=T.Vector3(-180.0, 180.0))
scale = NodeInput((1.0, 1.0, 1.0), ntype=T.Vector3(0.0, 5.0))
model_matrix = NodeOutput(ntype=T.Tensor((4, 4)))
def __init__(self, **kwargs):
super(ModelMatrix, self).__init__(**kwargs)
def update(self):
model = correctly_rotated_model_matrix()
glm.rotate(model, self.rotation[0], 1, 0, 0)
glm.rotate(model, self.rotation[1], 0, 1, 0)
glm.rotate(model, self.rotation[2], 0, 0, 1)
glm.translate(model, self.position[0], self.position[1], self.position[2])
glm.scale(model, self.scale[0], self.scale[1], self.scale[2])
self.model_matrix = model
class BinaryFunction(Node):
"""
Calculates the result of two input values x y and an arbitrary function string. It
is preferable to use the ComplexFunction Node as it provides additional values as
inputs and has better error checking.
Inputs:
x:
A float input.
y:
A float input.
function:
A function that should be computed at every timestep. Should be provided as
a string.
Outputs:
result:
Result of the computation.
"""
x = NodeInput(1.0, ntype=T.Float(-10, 10))
y = NodeInput(1.0, ntype=T.Float(-10, 10))
function = NodeInput("x+y", ntype=T.MathFunction())
result = NodeOutput(ntype=T.Float())
def update(self):
self.result = numexpr.evaluate(
self.function, local_dict={"x": self.x, "y": self.y}
)
class DrawNode(Node):
"""
The draw node is the standard final node in a graph to display something. All three
inputs must be connected.
Inputs:
faces:
A list of the faces that should be displayed. This needs to come from a
GeometryController Node as that node transmits the actual mesh data to the shader.
shader:
A ShaderProgram that runs the rendering pipeline.
window:
The window that the frame should be rendered on.
Outputs:
The DrawNode has no output, since it renders directly to the window.
"""
faces = NodeInput([0], ntype=T.List(T.UInt()))
shader = NodeInput(ntype=T.ShaderProgram())
window = NodeInput(ntype=T.Window())
def update(self):
# draw to screen
self.window.clear()
gl.glDisable(gl.GL_BLEND)
gl.glEnable(gl.GL_DEPTH_TEST)
gl.glEnable(gl.GL_POLYGON_OFFSET_FILL)
self.shader.draw(gl.GL_TRIANGLES, self.faces)
class Interpolate(Node):
"""
Interpolates between two vectors. The two vectors should have the same dimensions or
support pytorch broadcasting.
Inputs:
interpolation_value:
Where betweeen the two vectors the output should be set to.
Values between 0 and 1 make the most sense, but others are also valid and represent
an extrapolation.
input_vector0:
The first vector.
input_vector1:
The second vector.
Outputs:
result:
The interpolation result.
"""
interpolation_value = NodeInput(ntype=T.Float())
input_vector0 = NodeInput(torch.zeros(1), ntype=T.Tensor())
input_vector1 = NodeInput(torch.ones(1), ntype=T.Tensor())
result = NodeOutput(ntype=T.Tensor())
def update(self):
self.result = torch.lerp(
self.input_vector0, self.input_vector1, float(self.interpolation_value)
)
class Biggan(Node):
"""
The Biggan Node generates pictures based on a noise_vector and a class_vector(These
can be generated by the NoiseSampler/ClassSampler Nodes).
Inputs:
noise_vector:
A noise vector in the shape of (1, 128).
class_vector:
A class vector in the shape of (1, 1).
Outputs:
image:
A generated output image.
"""
noise_vector = NodeInput(
create_noise_vector(batch_size=1, truncation=1, device="cpu"), ntype=T.Tensor()
)
class_vector = NodeInput(
create_class_vector(643, batch_size=1, device="cpu"), ntype=T.Tensor()
)
image = NodeOutput(ntype=T.Tensor())
def __init__(self, model_type: str = "biggan-deep-128", **kwargs):
super(Biggan, self).__init__(**kwargs)
self.model = self.load_model(model_type, device=self.device)
self.truncation = torch.ones(1, device=self.device)
self.model_type = model_type
def update(self):
with torch.no_grad():
self.image = self.model(
self.noise_vector, self.class_vector, self.truncation
)
def to_device(self, device: str):
if device != self.device:
self.model = self.load_model(
self.model_type, device="cuda" if device.startswith("cuda") else device
)
super().to_device(device)
self.truncation = self.truncation.to(device)
def load_model(self, model_type, device):
file_name = str(get_asset_folder() / "networks" / (model_type + "-" + device))
try:
return torch.jit.load(file_name)
except (ValueError, RuntimeError):
print(
"Could not find",
file_name,
"\n Trying to create file(requires network connection)",
)
generate_model_file(model_type, device)
return torch.jit.load(file_name)
class SampleManager(Node):
"""
Manages the sampling and correct replacement of some variable that can be sampled,
such as noise or class.
Inputs:
time:
Some time value. If it is lower than the value of the previous frame,
it signals a beat.
sampler:
The sample function from which samples are drawn.
Outputs:
vector0:
The current output vector.
vector1:
The next output vector.
"""
time = NodeInput(ntype=T.Float())
sampler = NodeInput(ntype=T.Sampler())
vector0 = NodeOutput(ntype=T.Tensor())
vector1 = NodeOutput(ntype=T.Tensor())
def __init__(self, **kwargs):
super(SampleManager, self).__init__(**kwargs)
self.sample = iter(range(2))
self.inner_vec0 = next(self.sample)
self.inner_vec1 = next(self.sample)
self.vector0 = self.inner_vec0
self.vector1 = self.inner_vec1
self.refresh = True
self.internal_time = 0
def update(self):
if self.refresh:
self.recalculate()
self.refresh = False
if self.internal_time - self.time > 0.001:
self.inner_vec0 = self.inner_vec1
self.inner_vec1 = self.sampler()
self.vector0 = self.inner_vec0
self.vector1 = self.inner_vec1
self.internal_time = self.time
def recalculate(self):
self.inner_vec0 = self.sampler()
self.inner_vec1 = self.sampler()
self.vector0 = self.inner_vec0
self.vector1 = self.inner_vec1
def to_device(self, device):
super().to_device(device)
self.refresh = True
class Window(Node):
"""
This node exists for resizing and rendering purposes. It provides access to both the
window size and the window.
Outputs:
size:
The size of the window as a (width, height) tuple.
window:
The window object.
"""
size = NodeOutput(ntype=T.Tuple(T.UInt(), T.UInt()))
window = NodeOutput(ntype=T.Window())
def __init__(self, **kwargs):
super(Window, self).__init__(**kwargs)
self.internal_window = WindowManager.get_window()
self.window = self.internal_window
self.size = self.internal_window.width, self.internal_window.height
def update(self):
self.size = WindowManager.get_window_size()
class ClassPool(Node):
"""
A class pool node, which has additional frontend logic to make the class ids are
properly selectable.
Inputs:
class_pool:
The Classes that should be used for possible interpolation targets.
Outputs:
output_class_pool:
The value to access the class pool.
"""
class_pool = NodeInput(list(range(0, 999)), ntype=T.List(T.UInt()))
output_class_pool = NodeOutput(ntype=T.List(T.UInt()))
def __init__(self, **kwargs):
super(ClassPool, self).__init__(**kwargs)
self.output_class_pool = self.class_pool
def update(self):
self.output_class_pool = self.class_pool
class ScalarToVector(Node):
"""
Creates a vector2 from a scalar value
"""
scalar = NodeInput(0.0)
vector = NodeOutput(ntype=T.Vector2())
def __init__(self, vector_size=2, **kwargs):
super(ScalarToVector, self).__init__(**kwargs)
self.vector_size = vector_size
def update(self):
self.vector = [self.scalar] * self.vector_size
class NoiseSampler(Node):
"""
Samples noise as defined by the sampling scheme.
Inputs:
sample_options:
The mode how the noise will be sampled. All samples are drawn from a
normal distribution, if not specified. Possible modes are:
"PingPong":
Samples a vector and then inverts it on every consecutive sampling.
"SimpleRotation":
Samples the center(so all zeroes), except the 0th and first
dimension are sampled along a circle, the amount of rotation per sample is
given by the rotation_speed value.
"ComplexRotation":
Samples in a very chaotic rotation roughly around the center.
"Random":
Samples in a random manner.
"RandomStar":
Samples randomly, but every second sample is all zeroes.
truncation:
Multiplies the vector by a value, so the higher the truncation, the
higher the likelihood that the sample is nonsensical, the lower the truncation,
the smaller the sample variance.
rotation_speed:
Speed of rotation around the center for the two rotation sampling
modes. The higher, the faster the samples change.
Outputs:
sampler:
The sampling function. Every call creates a new sample according to the
sampling scheme.
"""
sample_options = NodeInput(
"Ping Pong",
ntype=T.Selector(
T.String(),
[
"Simple Rotation",
"Complex Rotation",
"Ping Pong",
"Random",
"Random Star",
],
),
)
truncation = NodeInput(1.0, ntype=T.Float(0, 10))
rotation_speed = NodeInput(0.125, ntype=T.Float(0, 1))
sampler = NodeOutput(ntype=T.Sampler())
def __init__(self, **kwargs):
super(NoiseSampler, self).__init__(**kwargs)
self.sampler = self.sample
self.star_center = self.truncation
self.step_counter = 1
self.class_pool = list(range(0, 5))
self.sample_schema = {
"Ping Pong": NoiseSampler.PingPong,
"Simple Rotation": NoiseSampler.SimpleRotation,
"Complex Rotation": NoiseSampler.ComplexRotation,
"Random": NoiseSampler.Random,
"RandomStar": NoiseSampler.RandomStar,
}
self.sample_mode = self.switch_mode()
self._in_edges["sample_options"].ntype.known_values = self.sample_schema.keys()
self.inner_option = self.sample_options
def update(self):
if self.inner_option != self.sample_options:
self.inner_option = self.sample_options
self.sample_mode = self.switch_mode()
def sample(self):
return self.sample_mode.generate(self.rotation_speed) * self.truncation
def switch_mode(self):
return self.sample_schema[self.sample_options](self.device)
def to_device(self, device):
super().to_device(device)
self.sample_mode = self.switch_mode()
class PingPong:
def __init__(self, device, **kwargs):
self.vector = create_noise_vector(
batch_size=1, dim_z=128, truncation=1, device=device,
)
def generate(self, rotation_speed):
self.vector = -self.vector
return self.vector
class SimpleRotation:
def __init__(self, device, **kwargs):
self.vector = torch.zeros((1, 128), device=device)
self.rotation = 0
def rotate(self, rotation_speed):
self.rotation += rotation_speed * math.pi
self.vector[:, 0] = sin(self.rotation)
self.vector[:, 1] = cos(self.rotation)
def generate(self, rotation_speed):
self.rotate(rotation_speed)
return self.vector
class ComplexRotation:
def __init__(self, device):
self.vector = torch.zeros((1, 128), device=device)
self.params = torch.zeros((64, 2), device=device).normal_()
self.rotation = 0
def rotate(self, rotation_speed):
self.rotation += rotation_speed * math.pi
for i, params in enumerate(self.params):
speed, distance = params
self.vector[:, 2 * i + 0] = distance * sin(speed * self.rotation)
self.vector[:, 2 * i + 1] = distance * cos(speed * self.rotation)
def generate(self, rotation_speed):
self.rotate(rotation_speed)
return self.vector
class Random:
def __init__(self, device):
self.device = device
def generate(self, rotation_speed):
return create_noise_vector(
batch_size=1, dim_z=128, truncation=1, device=self.device,
)
class RandomStar:
def __init__(self, device):
self.device = device
self.center = self.old_vector = (
create_noise_vector(
batch_size=1, dim_z=128, truncation=1, device=self.device,
)
* 0.0
)
self.center_next = False
def generate(self, rotation_speed):
if self.center_next:
vector = self.center
else:
vector = create_noise_vector(
batch_size=1, dim_z=128, truncation=1, device=self.device,
)
self.center_next = not self.center_next
return vector
class ClassSampler(Node):
"""
Samples classes as defined in the sampling scheme.
Inputs:
class_pool:
Pool of classes we sample from.
sample_options:
The sampling scheme that will be used. Currently avaiable schemes are: "star": Samples ray_count
nodes in consecutive order before resampling the center. The center is normally the class pool at index 0,
but can be changed with resample_center_every_n:
"continuous":
Samples the class_pool in round robin fashion.
"ping_pong":
Switches between the 0th and 1st entry of the class pool.
"random":
Samples a random class from the pool. Guarantees that no class while be
sampled 2 times in a row.
resample_center_every_n:
The bool decides if it the center in star mode should be resampled every n values,
the int is the n.
ray_count:
How many rays should be sampled before jumping back to the center.
Outputs:
sampler:
The sample function that can be used to sample a value.
"""
class_pool = NodeInput(list(range(0, 999)), ntype=T.List(T.UInt()))
sample_options = NodeInput(
"Random",
ntype=T.FixedSelector(
T.String(), ["Star", "Continuous", "Ping Pong", "Random"]
),
)
resample_center_every_n = NodeInput(
(False, 8), ntype=T.Tuple(T.Bool(), T.UInt(lower_bound=1, upper_bound=15))
)
ray_count = NodeInput(1, ntype=T.UInt(lower_bound=1, upper_bound=15))
sampler = NodeOutput(ntype=T.Sampler())
def __init__(self, **kwargs):
super(ClassSampler, self).__init__(**kwargs)
self.sampler = self.sample
self.sample_schema = {
"Star": ClassSampler.Star,
"Continuous": ClassSampler.Continuous,
"Ping Pong": ClassSampler.PingPong,
"Random": ClassSampler.Random,
}
self._in_edges["sample_options"].ntype.known_values = self.sample_schema.keys()
self.step_counter = 0
self.inner_option = self.sample_options
self.sample_mode = self.switch_mode()
def update(self):
if self.inner_option != self.sample_options:
self.inner_option = self.sample_options
self.sample_mode = self.switch_mode()
def sample(self):
self.step_counter += 1
if len(self.class_pool) == 1:
self.class_label = self.class_pool[0]
else:
self.class_label = self.sample_mode.generate(
self.class_pool,
self.step_counter,
self.resample_center_every_n,
self.ray_count,
)
return create_class_vector(self.class_label, batch_size=1, device=self.device)
def switch_mode(self):
return self.sample_schema[self.sample_options](self.class_pool)
class Random:
def __init__(self, class_pool):
self.class_label = -1
def generate(self, class_pool, *args, **kwargs):
class_label = self.class_label
while class_label == self.class_label:
self.class_label = choice(class_pool)
return self.class_label
class Star:
def __init__(self, class_pool):
self.star_center = self.sample_class_vector = choice(class_pool)
def generate(
self, class_pool, step_counter, resample_center_every_n, ray_count
):
resample, every_n = resample_center_every_n
if resample:
if step_counter % (every_n * (ray_count + 1)) == 0:
self.star_center = choice(class_pool)
else:
self.star_center = class_pool[0]
if step_counter % (ray_count + 1) == 0:
sample_class_vector = self.star_center
else:
sample_class_vector = class_pool[
1 + int(step_counter / 2) % (len(class_pool) - 1)
]
return sample_class_vector
class PingPong:
def __init__(self, class_pool):
pass
def generate(
self, class_pool, step_counter, resample_center_every_n, ray_count
):
return class_pool[step_counter % 2]
class Continuous:
def __init__(self, class_pool):
pass
def generate(
self, class_pool, step_counter, resample_center_every_n, ray_count
):
return class_pool[step_counter % len(class_pool)]
class RecordingDrawNode(Node):
"""
The recording draw node is the a final node in a graph to both display and record
something. All three inputs faces, shader and window must be connected.
Inputs:
faces:
A list of the faces that should be displayed. This needs to come from a
GeometryController Node as that node transmits the actual mesh data to the shader.
shader:
A ShaderProgram that runs the rendering pipeline.
window:
The window that the frame should be rendered on.
Outputs:
recording:
A bool if a video should be recorded or not.
song_name:
The song that should be the background music for the recorded video.
"""
faces = NodeInput([0], ntype=T.List(T.UInt()))
shader = NodeInput(ntype=T.ShaderProgram())
window = NodeInput(ntype=T.Window())
recording = NodeInput(False, ntype=T.Bool())
song_name = NodeInput("", ntype=T.FixedSelector(T.String(), song_dict.keys(),),)
def __init__(
self, video_folder=get_asset_folder() / "videos", **kwargs,
):
super(RecordingDrawNode, self).__init__(**kwargs)
self.currently_recording = False
self.video_folder = video_folder
self.fps = 15
self.recording_time = 0
def write_current_frame(self):
gl.glReadPixels(
0,
0,
self.window.width,
self.window.height,
gl.GL_RGB,
gl.GL_UNSIGNED_BYTE,
self.frame_buffer,
)
self.writer.write_frame(np.flipud(self.frame_buffer))
def update(self):
self.window.clear()
gl.glDisable(gl.GL_BLEND)
gl.glEnable(gl.GL_DEPTH_TEST)
gl.glEnable(gl.GL_POLYGON_OFFSET_FILL)
self.shader.draw(gl.GL_TRIANGLES, self.faces)
self.check_recording()
def check_recording(self):
if self.recording:
# start a new stream
if not self.currently_recording:
self.recording_time = 0
self.currently_recording = True
self.writer, self.frame_buffer = self.declare_writer()
if self.recording_time % self.fps == 0:
print(f"recording time: {int(self.recording_time / self.fps)}s")
self.recording_time += 1
self.write_current_frame()
else:
if self.currently_recording:
self.currently_recording = False
self.writer.close()
def declare_writer(self):
"""
Create a ffmpeg writer and define the output file name.
For this to work a change to the FFMPEG_VideoWriter backend has to be done:
- search and remove -an
- optional so the song works: add '-shortest' to the audio option
- https://github.com/Zulko/moviepy/pull/968 if merged maybe this is not needed anymore
Returns:
Ffmpeg writer and the frame buffer.
"""
shape = 512
try:
audio_file = song_dict[self.song_name]
except KeyError:
audio_file = None
from glumpy.ext.ffmpeg_writer import FFMPEG_VideoWriter # NOQA
self.output_file_name = (
str(self.video_folder)
+ f"/{self.song_name}_"
+ datetime.now().strftime("%d:%m_%H:%M")
+ ".mp4"
)
ffpmeg_vw = FFMPEG_VideoWriter(
filename=self.output_file_name,
size=(shape, shape),
fps=self.fps,
preset="veryfast",
codec="libx264", # h264
audiofile=audio_file,
)
frame_buffer = np.zeros((shape, shape, 3), dtype=np.uint8)
return ffpmeg_vw, frame_buffer
class ModifiedBpm(Node):
"""
The ModifiedBpm node is used to modify bpm speed and then provide the same outputs.
Inputs:
in_beat_time:
A BeatTimeMessage object to modifiy.
time_modifier:
The speed value by which the time should be changed.
Outputs:
out_beat_time:
An object that contains all information outputted by the node in one
place.
counter:
A counter of the beats, so 10 will mean that 10 beats have passed in total.
isBeat:
If the current frame is on a beat or not.
time:
Total time that has passed since startup.
normalized_time:
The time normalized by the bpm, so that each unit is equal to the
time passing between two beats. So 12.20 will be the value after 12 beats
passing and 20 % along the way to the next beat. Is equal to counter +
local_time.
delta_time:
The time passed between two frames.
local_time:
The percent of time passed between the last and next beat. Basically, it
is a value between 0 and 1 that describes how far we are distanced to both
on a time basis.
"""
in_beat_time = NodeInput(ntype=T.BeatTime())
time_modifier = NodeInput(1.0, ntype=T.Float(0, 10))
out_beat_time = NodeOutput(ntype=T.BeatTime())
counter = NodeOutput(0)
isBeat = NodeOutput(False)
time = NodeOutput(0)
delta_time = NodeOutput(0)
local_time = NodeOutput(0)
normalized_time = NodeOutput(0)
def __init__(self, **kwargs):
super().__init__(**kwargs)
self.last_local_time = 0
def update(self):
bpm = self.in_beat_time.bpm
time = self.in_beat_time.time * self.time_modifier
counter, local_time = divmod(time, 60 / bpm)
local_time *= bpm / 60
isBeat = self.last_local_time > local_time
self.last_local_time = local_time
delta_time = self.in_beat_time.delta_time * self.time_modifier
bt = BeatTimeMessage(
counter=counter,
isBeat=isBeat,
time=time,
delta_time=delta_time,
local_time=local_time,
bpm=bpm,
speed=self.time_modifier,
)
self.counter = counter
self.isBeat = isBeat
self.time = time
self.delta_time = delta_time
self.local_time = local_time
self.out_beat_time = bt
self.normalized_time = counter + local_time
class ComplexFunction(Node):
"""
A complex function with multiple possible inputs. Also allows slower playback by
changing the single_beat_length.
Inputs:
single_beat_length:
The length of a single beat. So if 1, the x value will change
from 0 to 1 over the course of a single beat. If 4, it will be 4 times as slow.
x:
An input time value in the normalized_time format. While other values could
be used, this is the de facto standard.
y:
An additional input value, can be bound to something if needed.
z:
An additional input value, can be bound to something if needed.
abc:
A tuple of 3 additional inputs.
function:
A function to compute.
Outputs:
result:
The result of computing the function. after changing the beat_length.
linear_result:
The result of just changing the beat_length with single_beat_length.
"""
single_beat_length = NodeInput(1.0, ntype=T.Base2Int(1, 512))
x = NodeInput(1.0, ntype=T.Float(0, 1))
y = NodeInput(1.0, ntype=T.Float(0, 1))
z = NodeInput(1.0, ntype=T.Float(0, 1))
abc = NodeInput((1.0, 1.0, 1.0), ntype=T.Vector3(0, 1))
function = NodeInput(
"x+y",
ntype=T.Selector(
T.MathFunction(),
[
"x",
"y",
"0",
"1",
"0.5-cos(x*3*pi)",
"x**y*(a)+(1-a)",
"x**y*(a+b*sin(c*x*pi))+(1-(a+b*sin(c*x*pi)))",
],
),
)
result = NodeOutput(ntype=T.Float())
linear_result = NodeOutput(ntype=T.Float())
def __init__(self, presets=None, bounds=(0, 1), **kwargs):
super(ComplexFunction, self).__init__(**kwargs)
if presets:
self._in_edges["function"].ntype.known_values += presets
self.working_function = self.function
self.bounds = bounds
def update(self):
sbl = max(0.01, self.single_beat_length)
x = math.fmod(self.x, sbl) / sbl
self.linear_result = x
try:
result = numexpr.evaluate(
self.function,
local_dict={
"x": x,
"y": self.y,
"z": self.z,
"a": self.abc[0],
"b": self.abc[1],
"c": self.abc[2],
"pi": math.pi,
},
)
except (KeyError, SyntaxError):
result = numexpr.evaluate(
self.working_function,
local_dict={
"x": x,
"y": self.y,
"z": self.z,
"a": self.abc[0],
"b": self.abc[1],
"c": self.abc[2],
"pi": math.pi,
},
)
self.result = np.clip(result, self.bounds[0], self.bounds[1])
class Shader(Node):
"""
The shader node manages the Vertex and Fragment Shader of the pipeline. It has
numerous inputs that manage different effects. It does not manage the geometry or
texture parts of the shader inputs, that is the mesh, its position or texture.
Inputs:
apply_sobel:
If True, applies edge detection through the sobel operator.
sobel_resolution:
The raster resolution for the sobel operator. The higher, the
finer the edge results.
is_mirrored:
If True, mirrors the texture along the (u,v) axes.
repeat_times:
How often the texture should be repeated in the (u,v) coordinate space.
texture_scaling_x:
Multiplies the texture coordinates in the u dimension.
texture_scaling_y:
Multiplies the texture coordinates in the v dimension.
is_crosshatched:
Applies the crosshatching effect on the texture.
is_inverted:
Inverts the output colors.
halftone_resolution:
If >0, applies the halftone effect. The greater the value, the
finer the resolution.
brightness:
Multiplies the pixel output by this value to control pixel brightness.
Outputs:
shader:
The shader program as an output, needed for both the GeometryController and
the DrawNode.
"""
apply_sobel = NodeInput(False, ntype=T.Bool())
sobel_resolution = NodeInput(512.0, ntype=T.Float(0.1, 1000))
is_mirrored = NodeInput((False, False), ntype=T.Tuple(T.Bool(), T.Bool()))
repeat_times = NodeInput(
[1.0, 1.0], ntype=T.Vector2(0, 10.0), pre_bind_hook=toNumpy
)
texture_scaling_x = NodeInput(1.0)
texture_scaling_y = NodeInput(1.0)
is_crosshatched = NodeInput(False, ntype=T.Bool())
is_inverted = NodeInput(False, ntype=T.Bool())
halftone_resolution = NodeInput(0.0, ntype=T.Float(0, 10))
brightness = NodeInput(1.0, ntype=T.Float(0.0, 1.0))
# transparency = NodeInput(1.0, ntype=T.Float(0.0, 1.0))
shader = NodeOutput(ntype=T.ShaderProgram())
def __init__(self, **kwargs):
super(Shader, self).__init__(**kwargs)
vertex = get_shader("vertex.glsl")
fragment = get_shader("fragment.glsl")
self.internal_shader = gloo.Program(vertex, fragment)
self.shader = self.internal_shader
def update(self):
self.internal_shader["apply_sobel"] = self.apply_sobel
self.internal_shader["u_resolution"] = self.sobel_resolution
self.internal_shader["mirror_x"] = self.is_mirrored[0]
self.internal_shader["mirror_y"] = self.is_mirrored[1]
self.internal_shader["repeat"] = self.repeat_times
self.internal_shader["texture_scale"] = np.array(
[self.texture_scaling_x, self.texture_scaling_y], dtype=float
)
self.internal_shader["is_crosshatch"] = self.is_crosshatched
self.internal_shader["is_inverted"] = self.is_inverted
self.internal_shader["halftone_resolution"] = self.halftone_resolution
self.internal_shader["brightness"] = self.brightness
self.internal_shader["transparency"] = 1 # self.transparency
self.shader = self.internal_shader
class GeometryController(Node):
"""
The geometry controller node is one of the key nodes to render something. It manages
mesh creation, texturing and mesh positioning.
Inputs:
shader:
A shader program that the mesh data will be transmitted to.
texture:
A texture that will be displayed on the mesh.
mesh:
The mesh that will be displayed. Currently all available meshes are in the
assets/meshes/ folder and are selectable in the frontend. You can access the path
list through mesh_dict.
view_distance:
The position of the camera. It is currently only movable in the z
direction.
window_size:
The size of the window, should be bound the output of a Window node.
model_matrix:
All parameters for the positioning, scaling and rotation of the mesh.
Currently there are no extra measures against gimbal locking.
Outputs:
faces:
This output is a list of face ids that should be displayed. It should be
connected to a faces input on one of the recording nodes.
"""
shader = NodeInput(ntype=T.ShaderProgram())
texture = NodeInput(torch.zeros(1, 3, 1, 1), ntype=T.Image())
mesh = NodeInput("quad", ntype=T.FixedSelector(T.String(), mesh_dict.keys(),),)
view_distance = NodeInput(-2.5, ntype=T.Float(-30, 5))
window_size = NodeInput(ntype=T.Tuple(T.UInt(), T.UInt()))
model_matrix = NodeInput(correctly_rotated_model_matrix(), ntype=T.Tensor((4, 4)))
faces = NodeOutput(ntype=T.List(T.UInt()))
def __init__(self, **kwargs):
super(GeometryController, self).__init__(**kwargs)
self.internal_mesh = ""
def update(self):
if self.mesh != self.internal_mesh:
self.load_mesh(mesh_dict[self.mesh])
self.internal_mesh = self.mesh
# The model is rotated because the texturing is the wrong way around
self.shader["model"] = self.model_matrix
self.shader["view"] = glm.translation(0, 0, self.view_distance)
# Scale is increased to fill the whole screen
self.shader["scale"] = 1.0
self.shader["tex"] = to_tex_format(self.texture)
width, height = self.window_size
self.shader["projection"] = glm.perspective(
# 45.0, 1.0, 2.0, 100.0
45.0,
width / float(height),
2.0,
100.0,
)
def load_mesh(self, file):
vertices, faces = load_file(file)
self.faces = faces
self.shader.bind(vertices)
self.has_update = False
class BpmClock(Node):
"""
The BpmClock is one of the core nodes for a standard Rendergraph. It gives out
multiple signals synchronized to a bpm, typically the bpm of the song that is
currently playing.
Inputs:
bpm:
A beats per minute value.
Outputs:
beat_time:
An object that contains all information outputted by the node in one
place.
counter:
A counter of the beats, so 10 will mean that 10 beats have passed in total.
isBeat:
If the current frame is on a beat or not.
time:
Total time that has passed since startup.
normalized_time:
The time normalized by the bpm, so that each unit is equal to the
time passing between two beats. So 12.20 will be the value after 12 beats passing and
20 % along the way to the next beat. Is equal to counter + local_time.
delta_time:
The time passed between two frames.
local_time:
The percent of time passed between the last and next beat. Basically, it
is a value between 0 and 1 that describes how far we are distanced to both on a time
basis.
"""
bpm = NodeInput(128.0, ntype=T.Float(lower_bound=0.00001, upper_bound=300))
beat_time = NodeOutput(BeatTimeMessage(), ntype=T.BeatTime())
counter = NodeOutput(0)
isBeat = NodeOutput(False)
time = NodeOutput(0.0)
normalized_time = NodeOutput(0.0)
delta_time = NodeOutput(0.0)
local_time = NodeOutput(0.0)
def __init__(self, time_generator=time_generator(), **kwargs):
super().__init__(**kwargs)
self.time_gen = time_generator
self.internal_time = 0
self.internal_counter = 0
self.total_time = 0
self.init_time = next(self.time_gen)
self.last_time = self.init_time
self.last_local_time = 0
def update(self):
time = next(self.time_gen) - self.init_time
counter, local_time = divmod(time, 60 / self.bpm)
local_time *= self.bpm / 60
isBeat = self.last_local_time > local_time
self.last_local_time = local_time
delta_time = time - self.last_time
self.last_time = time
bt = BeatTimeMessage(
counter=counter,
isBeat=isBeat,
time=time,
delta_time=delta_time,
local_time=local_time,
bpm=self.bpm,
speed=1,
)
self.counter = counter
self.isBeat = isBeat
self.time = time
self.delta_time = delta_time
self.local_time = local_time
self.beat_time = bt
self.normalized_time = local_time + counter