def write_out_vel(particles,
vel,
d,
v,
t,
suffix,
xlim=None,
ylim=None,
s=5,
z=None):
writeNumpyRaw((dst_path + suffix + "_%03d") % (d, v, t), vel)
plot_particles(particles,
xlim,
ylim,
s, (dst_path + suffix + "_%03d.png") % (d, v, t),
src=particles,
vel=vel,
z=z)
plot_particles(particles,
xlim,
ylim,
s, (dst_path + suffix + "_%03d.svg") % (d, v, t),
src=particles,
vel=vel,
z=z)
write_csv((dst_path + suffix + "_%03d.csv") % (d, v, t), vel)
def add_images(writer,
tag,
src,
ref,
res,
epoch,
xlim=None,
ylim=None,
s=1.0,
z=None):
imgs = []
buf = io.BytesIO()
plot_particles(res,
path=buf,
src=src,
ref=ref,
xlim=xlim,
ylim=ylim,
s=s,
z=z)
img = tf.Summary.Image(colorspace=4, encoded_image_string=buf.getvalue())
imgs.append(tf.Summary.Value(tag=tag + "_comp", image=img))
buf = io.BytesIO()
plot_particles(res, path=buf, xlim=xlim, ylim=ylim, s=s, z=z)
img = tf.Summary.Image(colorspace=4, encoded_image_string=buf.getvalue())
imgs.append(tf.Summary.Value(tag=tag + "_res", image=img))
buf = io.BytesIO()
plot_particles(src, path=buf, xlim=xlim, ylim=ylim, s=s, z=z)
img = tf.Summary.Image(colorspace=4, encoded_image_string=buf.getvalue())
imgs.append(tf.Summary.Value(tag=tag + "_src", image=img))
buf = io.BytesIO()
plot_particles(ref, path=buf, xlim=xlim, ylim=ylim, s=s, z=z)
img = tf.Summary.Image(colorspace=4, encoded_image_string=buf.getvalue())
imgs.append(tf.Summary.Value(tag=tag + "_ref", image=img))
writer.add_summary(tf.Summary(value=imgs), epoch)
writer.flush()
return
0 if len(patch_pos) == 3 else 2,
aux_data=par_aux,
features=features,
pad_val=pad_val,
bnd=bnd,
last_pos=positions,
stride_hys=1.0)
if len(patch_pos) == 3:
idx = get_nearest_idx(patch_extractor.positions, patch_pos)
patch = patch_extractor.get_patch(idx, False)
plot_particles(patch_extractor.positions,
[0, int(out_res / factor_d[0])],
[0, int(out_res / factor_d[0])],
5,
tmp_path + "patch_centers_%03d.png" % i,
np.array([patch_extractor.positions[idx]]),
np.array([patch_pos]),
z=patch_pos[2] if dim == 3 else None)
patch_pos = patch_extractor.positions[idx] + par_aux['v'][
patch_extractor.pos_idx[idx]] / data_config['fps']
ref_patch = extract_particles(ref_data[i], patch_pos * factor_d,
par_cnt_dst, half_ps, pad_val)[0]
result = eval_patch(punet, [np.array([patch])],
tmp_path + "result_%s" + "_%03d" % i,
z=None if dim == 2 else 0,
verbose=3 if verbose else 1)
hdr = OrderedDict([('dim', len(result)), ('dimX', int(patch_size_ref)),
('dimY', int(patch_size_ref)),
('dimZ', 1 if dim == 2 else int(patch_size_ref)),
1,
name="coord_reconstruction_1",
activation=activation)(x)
b = np.zeros((3, ), dtype='float32')
W = np.zeros((1, fac * 8, 3), dtype='float32')
x = Conv1D(3, 1, name="coord_reconstruction_2", activation=activation)(x)
x = Reshape((100, 3))(x)
m = Model(inputs=inputs, outputs=x)
m.compile(
keras.optimizers.adam(),
loss=lambda y_true, y_pred: repulsion_loss(y_pred, 20, 0.1, 0.005))
data = np.random.random((10000, 100, 3)) * 0.1
plot_particles(m.predict(data[:1])[0], src=data[0], s=5)
m.fit(data, data, epochs=1)
plot_particles(m.predict(data[:1])[0], src=data[0], s=5)
m.fit(data, data, epochs=1)
plot_particles(m.predict(data[:1])[0], src=data[0], s=5)
m.fit(data, data, epochs=1)
plot_particles(m.predict(data[:1])[0], src=data[0], s=5)
m.fit(data, data, epochs=1)
plot_particles(m.predict(data[:1])[0], src=data[0], s=5)
m.fit(data, data, epochs=1)
plot_particles(m.predict(data[:1])[0], src=data[0], s=5)
src_grid.sample_sdf()
ref_grid.sample_sdf()
return src_grid, ref_grid
if __name__ == '__main__':
from neuralparticles.tools.plot_helpers import plot_sdf, plot_particles
from neuralparticles.tools.data_helpers import extract_particles
rp = RandomParticles(50, 50, 1, 2, 3, 10, 5, 0.5, 3)
rp.gen_random()
src_grid, ref_grid = rp.get_grid()
plot_particles(src_grid.particles, [0, src_grid.dimX], [0, src_grid.dimY],
0.1, "src.pdf")
plot_particles(ref_grid.particles, [0, ref_grid.dimX], [0, ref_grid.dimY],
0.1, "ref.pdf")
plot_particles(src_grid.particles, [0, src_grid.dimX], [0, src_grid.dimY],
0.1, "comp.pdf", ref_grid.particles / 3)
plot_sdf(src_grid.cells[0, :, :, 0], [0, src_grid.dimX],
[0, src_grid.dimY],
"src_sdf.pdf",
src=src_grid.particles,
s=0.1)
plot_sdf(ref_grid.cells[0, :, :, 0], [0, ref_grid.dimX],
[0, ref_grid.dimY],
"ref_sdf.pdf",
src=ref_grid.particles,
s=0.1)
def eval_frame(model,
patch_extractor,
factor_d,
path="",
src=None,
aux=None,
ref=None,
hdim=0,
z=None,
verbose=0):
patches = patch_extractor.get_patches()
result = model.predict(patches)
if patch_extractor.data.shape[0] > 0:
if z is None:
tmp = np.repeat(patch_extractor.data, factor_d**2, axis=0)
displace = np.transpose(
np.reshape(np.mgrid[:factor_d, :factor_d, :1] + 0.5,
(3, -1))) / factor_d - 0.5
displace = np.concatenate(
np.repeat(np.expand_dims(displace, axis=0),
patch_extractor.data.shape[0],
axis=0))
patch_extractor.data = tmp + displace * np.array(
[1., 1., 0.]
) # + np.random.normal(scale=1/(3*factor_d), size=displace.shape)) * np.array([1.,1.,0.])
else:
tmp = np.repeat(patch_extractor.data, factor_d**3, axis=0)
displace = np.transpose(
np.reshape(np.mgrid[:factor_d, :factor_d, :factor_d] + 0.5,
(3, -1))) / factor_d - 0.5
displace = np.concatenate(
np.repeat(np.expand_dims(displace, axis=0),
patch_extractor.data.shape[0],
axis=0))
patch_extractor.data = tmp + displace # + np.random.normal(scale=1/(3*factor_d), size=displace.shape)
if type(result) is list:
for i in range(len(patch_extractor.positions)):
patch_extractor.set_patch(
result[0][i, :int(result[1][i] * result[0].shape[1])], i)
else:
for i in range(len(patch_extractor.positions)):
cnt = int(np.count_nonzero(patches[0][..., 1] != -2.0)) * (
result.shape[1] // patches[0].shape[1])
patch_extractor.set_patch(result[i, :cnt], i)
result = patch_extractor.data * np.array(
[factor_d, factor_d, 0 if z is None else factor_d])
if path != "" and verbose > 0:
vel_src = None
'''for k in aux:
aux_src = aux[k]
if k == 'v':
vel_src = aux_src/100
if verbose > 2: write_csv(path + "_%s.csv"%k, aux_src)'''
if verbose > 0:
plot_particles(result,
xlim=[0, hdim],
ylim=[0, hdim],
s=0.1,
path=path % ("comp") + ".png",
ref=ref,
src=src * factor_d,
vel=vel_src,
z=z)
plot_particles(src,
xlim=[0, hdim // factor_d],
ylim=[0, hdim // factor_d],
s=0.1,
path=path % ("src") + ".png",
src=src,
vel=vel_src,
z=z)
if ref is not None:
plot_particles(ref,
xlim=[0, hdim],
ylim=[0, hdim],
s=0.1,
path=path % ("ref") + ".png",
z=z)
plot_particles(result,
xlim=[0, hdim],
ylim=[0, hdim],
s=0.1,
path=path % ("res") + ".png",
z=z)
if verbose > 1:
plot_particles(result,
xlim=[0, hdim],
ylim=[0, hdim],
s=0.1,
path=path % ("comp") + ".svg",
ref=ref,
src=src * factor_d,
vel=vel_src,
z=z)
plot_particles(src,
xlim=[0, hdim // factor_d],
ylim=[0, hdim // factor_d],
s=0.1,
path=path % ("src") + ".svg",
src=src,
vel=vel_src,
z=z)
if ref is not None:
plot_particles(ref,
xlim=[0, hdim],
ylim=[0, hdim],
s=0.1,
path=path % ("ref") + ".svg",
z=z)
plot_particles(result,
xlim=[0, hdim],
ylim=[0, hdim],
s=0.1,
path=path % ("res") + "svg",
z=z)
if verbose > 2:
write_csv(path % ("res") + ".csv", result)
if ref is not None: write_csv(path % ("ref") + ".csv", ref)
write_csv(path % ("src") + ".csv", src)
patch_extractor.reset()
return result
def eval_patch(model,
src,
path="",
ref=None,
features=[],
z=None,
verbose=0,
truncate=False):
result = model.predict(src)
if type(result) is list:
raw_result = result[0][0]
cnt = int(result[1][0] * result[0].shape[1])
if verbose > 0: print("Reduce points to: " + str(cnt))
result = result[0][0, :cnt]
elif truncate:
raw_result = result[0]
cnt = int(np.count_nonzero(
src[0][..., 1] != -2.0)) * (result.shape[1] // src[0].shape[1])
if verbose > 0: print("Reduce points to: " + str(cnt))
result = result[0, :cnt]
else:
raw_result = result[0]
if verbose > 0: print("No Truncation!")
result = result[0]
if path != "" and verbose > 0:
_i = 0
vel_src = None
'''for i in range(len(features)):
if features[i] == 'v':
aux_src = src[1][:,_i:_i+3]
vel_src = aux_src/100
_i += 3
else:
aux_src = src[1][:,_i:_i+1]
_i += 1
if verbose > 2: write_csv(path + "_%s.csv"%features[i], aux_src)'''
if verbose > 0:
permutate = True
print(
cluster_analysis(src[0][0][..., :3], raw_result,
result.shape[0], permutate))
"""c = np.arange(src[0][0].shape[0])
#c[0] = 0
if permutate:
c = np.repeat(c, raw_result.shape[0]//src[0][0].shape[0])[:result.shape[0]]
else:
c = np.repeat(np.expand_dims(c,0), raw_result.shape[0]//src[0][0].shape[0]).flatten()[:result.shape[0]]
plot_particles(result, xlim=[-1,1], ylim=[-1,1], s=5, path=path%("detail") + ".png", src=src[0][0], vel=vel_src, z = z, c=c)"""
fac = raw_result.shape[0] // src[0][0].shape[0]
plot_particles(result,
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("comp") + ".png",
ref=ref,
src=src[0][0],
vel=vel_src,
z=z)
plot_particles(result[fac:],
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("detail") + ".png",
src=src[0][0],
vel=vel_src,
z=z,
ref=result[:fac])
plot_particles(src[0][0],
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("src") + ".png",
src=src[0][0],
vel=vel_src,
z=z)
if ref is not None:
plot_particles(ref,
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("ref") + ".png",
z=z)
plot_particles(result,
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("res") + ".png",
z=z)
if verbose > 1:
plot_particles(result,
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("comp") + ".svg",
ref=ref,
src=src[0][0],
vel=vel_src,
z=z)
plot_particles(result[fac:],
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("detail") + ".svg",
src=src[0][0],
vel=vel_src,
z=z,
ref=result[:fac])
plot_particles(src[0][0],
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("src") + ".svg",
src=src[0][0],
vel=vel_src,
z=z)
if ref is not None:
plot_particles(ref,
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("ref") + ".svg",
z=z)
plot_particles(result,
xlim=[-1, 1],
ylim=[-1, 1],
s=5,
path=path % ("res") + ".svg",
z=z)
if verbose > 2:
write_csv(path % ("res") + ".csv", result)
if ref is not None: write_csv(path % ("ref") + ".csv", ref)
write_csv(path % ("src") + ".csv", src[0][0])
return result