/
plot_latent_vs_true.py
236 lines (199 loc) · 8.76 KB
/
plot_latent_vs_true.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import torch
from torch.autograd import Variable
from torch.utils.data import DataLoader
import brewer2mpl
bmap = brewer2mpl.get_map('Set1', 'qualitative', 3)
colors = bmap.mpl_colors
plt.style.use('ggplot')
VAR_THRESHOLD = 1e-2
def plot_vs_gt_shapes(vae, shapes_dataset, save, z_inds=None):
dataset_loader = DataLoader(shapes_dataset, batch_size=1000, num_workers=1, shuffle=False)
N = len(dataset_loader.dataset) # number of data samples
K = vae.z_dim # number of latent variables
nparams = vae.q_dist.nparams
vae.eval()
# print('Computing q(z|x) distributions.')
qz_params = torch.Tensor(N, K, nparams)
n = 0
for xs in dataset_loader:
batch_size = xs.size(0)
xs = Variable(xs.view(batch_size, 1, 64, 64).cuda(), volatile=True)
qz_params[n:n + batch_size] = vae.encoder.forward(xs).view(batch_size, vae.z_dim, nparams).data
n += batch_size
qz_params = qz_params.view(3, 6, 40, 32, 32, K, nparams)
# z_j is inactive if Var_x(E[z_j|x]) < eps.
qz_means = qz_params[:, :, :, :, :, :, 0]
var = torch.std(qz_means.contiguous().view(N, K), dim=0).pow(2)
active_units = torch.arange(0, K)[var > VAR_THRESHOLD].long()
print('Active units: ' + ','.join(map(str, active_units.tolist())))
n_active = len(active_units)
print('Number of active units: {}/{}'.format(n_active, vae.z_dim))
if z_inds is None:
z_inds = active_units
# subplots where subplot[i, j] is gt_i vs. z_j
mean_scale = qz_means.mean(2).mean(2).mean(2) # (shape, scale, latent)
mean_rotation = qz_means.mean(1).mean(2).mean(2) # (shape, rotation, latent)
mean_pos = qz_means.mean(0).mean(0).mean(0) # (pos_x, pos_y, latent)
fig = plt.figure(figsize=(3, len(z_inds))) # default is (8,6)
gs = gridspec.GridSpec(len(z_inds), 3)
gs.update(wspace=0, hspace=0) # set the spacing between axes.
vmin_pos = torch.min(mean_pos)
vmax_pos = torch.max(mean_pos)
for i, j in enumerate(z_inds):
ax = fig.add_subplot(gs[i * 3])
ax.imshow(mean_pos[:, :, j].numpy(), cmap=plt.get_cmap('coolwarm'), vmin=vmin_pos, vmax=vmax_pos)
ax.set_xticks([])
ax.set_yticks([])
ax.set_ylabel(r'$z_' + str(j) + r'$')
if i == len(z_inds) - 1:
ax.set_xlabel(r'pos')
vmin_scale = torch.min(mean_scale)
vmax_scale = torch.max(mean_scale)
for i, j in enumerate(z_inds):
ax = fig.add_subplot(gs[1 + i * 3])
ax.plot(mean_scale[0, :, j].numpy(), color=colors[2])
ax.plot(mean_scale[1, :, j].numpy(), color=colors[0])
ax.plot(mean_scale[2, :, j].numpy(), color=colors[1])
ax.set_ylim([vmin_scale, vmax_scale])
ax.set_xticks([])
ax.set_yticks([])
x0, x1 = ax.get_xlim()
y0, y1 = ax.get_ylim()
ax.set_aspect(abs(x1 - x0) / abs(y1 - y0))
if i == len(z_inds) - 1:
ax.set_xlabel(r'scale')
vmin_rotation = torch.min(mean_rotation)
vmax_rotation = torch.max(mean_rotation)
for i, j in enumerate(z_inds):
ax = fig.add_subplot(gs[2 + i * 3])
ax.plot(mean_rotation[0, :, j].numpy(), color=colors[2])
ax.plot(mean_rotation[1, :, j].numpy(), color=colors[0])
ax.plot(mean_rotation[2, :, j].numpy(), color=colors[1])
ax.set_ylim([vmin_rotation, vmax_rotation])
ax.set_xticks([])
ax.set_yticks([])
x0, x1 = ax.get_xlim()
y0, y1 = ax.get_ylim()
ax.set_aspect(abs(x1 - x0) / abs(y1 - y0))
if i == len(z_inds) - 1:
ax.set_xlabel(r'rotation')
fig.text(0.5, 0.03, 'Ground Truth', ha='center')
fig.text(0.01, 0.5, 'Learned Latent Variables ', va='center', rotation='vertical')
plt.savefig(save)
plt.close()
def plot_vs_gt_faces(vae, faces_dataset, save, z_inds=None):
dataset_loader = DataLoader(faces_dataset, batch_size=1000, num_workers=1, shuffle=False)
N = len(dataset_loader.dataset) # number of data samples
K = vae.z_dim # number of latent variables
nparams = vae.q_dist.nparams
vae.eval()
# print('Computing q(z|x) distributions.')
qz_params = torch.Tensor(N, K, nparams)
n = 0
for xs in dataset_loader:
batch_size = xs.size(0)
xs = Variable(xs.view(batch_size, 1, 64, 64).cuda(), volatile=True)
qz_params[n:n + batch_size] = vae.encoder.forward(xs).view(batch_size, vae.z_dim, nparams).data
n += batch_size
qz_params = qz_params.view(50, 21, 11, 11, K, nparams)
# z_j is inactive if Var_x(E[z_j|x]) < eps.
qz_means = qz_params[:, :, :, :, :, 0]
var = torch.std(qz_means.contiguous().view(N, K), dim=0).pow(2)
active_units = torch.arange(0, K)[var > VAR_THRESHOLD].long()
print('Active units: ' + ','.join(map(str, active_units.tolist())))
n_active = len(active_units)
print('Number of active units: {}/{}'.format(n_active, vae.z_dim))
if z_inds is None:
z_inds = active_units
# subplots where subplot[i, j] is gt_i vs. z_j
mean_pose_az = qz_means.mean(3).mean(2).mean(0) # (pose_az, latent)
mean_pose_el = qz_means.mean(3).mean(1).mean(0) # (pose_el, latent)
mean_light_az = qz_means.mean(2).mean(1).mean(0) # (light_az, latent)
fig = plt.figure(figsize=(len(z_inds), 3)) # default is (8,6)
gs = gridspec.GridSpec(3, len(z_inds))
gs.update(wspace=0, hspace=0) # set the spacing between axes.
vmin_scale = torch.min(mean_pose_az)
vmax_scale = torch.max(mean_pose_az)
for i, j in enumerate(z_inds):
ax = fig.add_subplot(gs[i])
ax.plot(mean_pose_az[:, j].numpy())
ax.set_ylim([vmin_scale, vmax_scale])
ax.set_xticks([])
ax.set_yticks([])
x0, x1 = ax.get_xlim()
y0, y1 = ax.get_ylim()
ax.set_aspect(abs(x1 - x0) / abs(y1 - y0))
if i == 0:
ax.set_ylabel(r'azimuth')
vmin_scale = torch.min(mean_pose_el)
vmax_scale = torch.max(mean_pose_el)
for i, j in enumerate(z_inds):
ax = fig.add_subplot(gs[len(z_inds) + i])
ax.plot(mean_pose_el[:, j].numpy())
ax.set_ylim([vmin_scale, vmax_scale])
ax.set_xticks([])
ax.set_yticks([])
x0, x1 = ax.get_xlim()
y0, y1 = ax.get_ylim()
ax.set_aspect(abs(x1 - x0) / abs(y1 - y0))
if i == 0:
ax.set_ylabel(r'elevation')
vmin_scale = torch.min(mean_light_az)
vmax_scale = torch.max(mean_light_az)
for i, j in enumerate(z_inds):
ax = fig.add_subplot(gs[2 * len(z_inds) + i])
ax.plot(mean_light_az[:, j].numpy())
ax.set_ylim([vmin_scale, vmax_scale])
ax.set_xticks([])
ax.set_yticks([])
x0, x1 = ax.get_xlim()
y0, y1 = ax.get_ylim()
ax.set_aspect(abs(x1 - x0) / abs(y1 - y0))
if i == 0:
ax.set_ylabel(r'lighting')
plt.suptitle('GT Factors vs. Latent Variables')
plt.savefig(save)
plt.close()
if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('-checkpt', required=True)
parser.add_argument('-zs', type=str, default=None)
parser.add_argument('-gpu', type=int, default=0)
parser.add_argument('-save', type=str, default='latent_vs_gt.pdf')
parser.add_argument('-elbo_decomp', action='store_true')
args = parser.parse_args()
from elbo_decomposition import elbo_decomposition
import lib.dist as dist
import lib.flows as flows
from vae_quant import VAE, setup_data_loaders
def load_model_and_dataset(checkpt_filename):
print('Loading model and dataset.')
checkpt = torch.load(checkpt_filename, map_location=lambda storage, loc: storage)
args = checkpt['args']
state_dict = checkpt['state_dict']
# model
if not hasattr(args, 'dist') or args.dist == 'normal':
prior_dist = dist.Normal()
q_dist = dist.Normal()
elif args.dist == 'laplace':
prior_dist = dist.Laplace()
q_dist = dist.Laplace()
elif args.dist == 'flow':
prior_dist = flows.FactorialNormalizingFlow(dim=args.latent_dim, nsteps=4)
q_dist = dist.Normal()
vae = VAE(z_dim=args.latent_dim, use_cuda=True, prior_dist=prior_dist, q_dist=q_dist, conv=args.conv)
vae.load_state_dict(state_dict, strict=False)
# dataset loader
loader = setup_data_loaders(args)
return vae, loader, args
z_inds = list(map(int, args.zs.split(','))) if args.zs is not None else None
torch.cuda.set_device(args.gpu)
vae, dataset_loader, cpargs = load_model_and_dataset(args.checkpt)
if args.elbo_decomp:
elbo_decomposition(vae, dataset_loader)
eval('plot_vs_gt_' + cpargs.dataset)(vae, dataset_loader.dataset, args.save, z_inds)