def __init__(self, den_path, fdc_path):
self.den_model_pth = den_path
self.den = DEN()
self.den.load_state_dict(torch.load(self.den_model_pth), strict=False)
self.fdc = FDC(self.den)
self.fdc.load_weights(fdc_path)
self.crop_ratios = [0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95, 1]
self.transform = Compose([
transforms_nyu.Normalize(),
transforms_nyu.FDCPreprocess(self.crop_ratios)
])
class FDCPredictor:
def __init__(self, den_path, fdc_path):
self.den_model_pth = den_path
self.den = DEN()
self.den.load_state_dict(torch.load(self.den_model_pth), strict=False)
self.fdc = FDC(self.den)
self.fdc.load_weights(fdc_path)
self.crop_ratios = [0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95, 1]
self.transform = Compose([
transforms_nyu.Normalize(),
transforms_nyu.FDCPreprocess(self.crop_ratios)
])
def prediction(self, img_path):
print(img_path.split("."))
with open(img_path, 'rb') as f_img:
image = pickle.load(f_img)
img = Image.fromarray(image, 'RGB')
img.save(img_path + "ng")
print(img_path + "ng" + " saved!")
nyu_dict = {'image': image, 'depth': image}
cropped = self.transform(nyu_dict)['stacked_images']
cropped = cropped.unsqueeze(0)
bsize, crops, c, h, w = cropped.size()
print(bsize, crops, c, h, w)
return self.fdc(cropped)[0]
def save(self, img, des_path):
img = Image.fromarray(img.numpy())
img.save("/root/DEN/images/depth_img/" + des_path.split(".")[-2] +
".tiff")
print("/root/DEN/images/depth_img/" + des_path.split(".")[-2] +
".tiff" + "saved!")
def compute_linkage_matrix(model):
from fdc import FDC
"""Wrapper for constructing the final linkage matrix using build_dendrogram function
Parameters
---------------
model : FDC class object
Contains the coarse graining information determined by fitting and coarse graining
with the data
Return
-------
Z : linkage matrix - (n_coarse_grain, 4)
From scipy's definition : "An (n−1)(n−1) by 4 matrix Z is
returned. At the ii-th iteration, clusters with indices Z[i, 0]
and Z[i, 1] are combined to form cluster n+in+i.
A cluster with an index less than nn corresponds to one of
the nn original observations. The distance between clusters
Z[i, 0] and Z[i, 1] is given by Z[i, 2]. The fourth value
Z[i, 3] represents the number of original observations in
the newly formed cluster."
"""
from copy import deepcopy
assert type(model) == type(FDC()), 'wrong type !'
hierarchy = deepcopy(model.hierarchy)
noise_range = deepcopy(model.noise_range)
# ---- PADDING trick --- for plotting purposes ...
n_elem = len(hierarchy[-1]['cluster_labels'])
terminal_cluster = hierarchy[-1]['idx_centers'][0]
hierarchy.append({'idx_centers': [terminal_cluster], 'cluster_labels' : np.zeros(n_elem,dtype=int)})
noise_range.append(1.5*model.max_noise)
# -------------------------------------------
# -------------------------------------------
Z = build_dendrogram(hierarchy, noise_range)
return Z
'''
Created on Feb 1, 2017
@author: Alexandre Day
Purpose:
Perform density clustering on gaussian mixture
'''
from fdc import FDC
from sklearn.datasets import make_blobs
from fdc import plotting
import pickle
import numpy as np
n_true_center = 15
np.random.seed(0)
print("------> Example with %i true cluster centers <-------"%n_true_center)
X, y = make_blobs(10000, 2, n_true_center) # Generating random gaussian mixture
model = FDC(noise_threshold=0.05, nh_size=40) # specifying density clustering parameters
model.fit(X) # performing the clustering
plotting.set_nice_font() # nicer plotting font !
plotting.summary_model(model, ytrue=y, show=True, savefile="result.png")
from fdc import plotting
import pickle
import numpy as np
from matplotlib import pyplot as plt
n_true_center = 15
np.random.seed(0)
print("------> Example with %i true cluster centers <-------"%n_true_center)
X, y = make_blobs(10000, 2, n_true_center) # Generating random gaussian mixture
X = StandardScaler().fit_transform(X) # always normalize your data :)
# set eta=0.0 if you have excellent density profile fit (lots of data say)
model = FDC(eta = 0.01)#, atol=0.0001, rtol=0.0001)
model.fit(X) # performing the clustering
x = np.linspace(-0.5, 0.6,200)
y = 1.5*x+0.15
X_2 = np.vstack([x,y]).T
xy2 = X_2[65]
b=xy2[0]/1.5+xy2[1]
y2 = -x/1.5+b
#rho = np.exp(model.density_model.evaluate_density(X_2))
#plt.plot(rho)
#plt.show()
#exit()
plt.scatter(x, y, c="green", zorder=2)
"""
Setting FDC parameters (note these are the same across all datasets)
"""
noise_threshold = 1.0
datasets = [noisy_circles, noisy_moons, varied, aniso, blobs, no_structure]
for i_dataset, dataset in enumerate(datasets):
X, y = dataset
# normalize dataset for easier parameter selection
X = StandardScaler().fit_transform(X)
# create clustering estimators
model = FDC(noise_threshold=noise_threshold)
s=time.time()
model.fit(X)
dt=time.time()-s
n_center=len(model.idx_centers)
plt.subplot(3,2,plot_num)
plt.scatter(X[:, 0], X[:, 1], color=colors[model.cluster_label].tolist(), s=10,zorder=1)
plt.text(.99, .07, ('%.2fs' % (dt)).lstrip('0'),
transform=plt.gca().transAxes, size=15,
horizontalalignment='right',zorder=2,
#################################
#################################
#################################
"""
Setting FDC parameters (note these are the same across all datasets)
"""
datasets = [noisy_circles, noisy_moons, varied, aniso, blobs, no_structure]
for i_dataset, dataset in enumerate(datasets):
X, y = dataset
# normalize dataset for easier parameter selection
X = StandardScaler().fit_transform(X)
# create clustering estimators
# atol and rtol set the precision of the density map, higher value improves performanc but reduces accuracy
model = FDC(eta=0.4)
s = time.time()
model.fit(X)
dt = time.time() - s
n_center = len(model.idx_centers)
plt.subplot(3, 2, plot_num)
plt.scatter(X[:, 0],
X[:, 1],
color=colors[model.cluster_label].tolist(),
s=10,
zorder=1)