forked from gr8joo/MVTCAE
-
Notifications
You must be signed in to change notification settings - Fork 0
/
plotting.py
164 lines (142 loc) · 7.03 KB
/
plotting.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
import sys
import os
import numpy as np
import torch
import torch.nn as nn
from torch.autograd import Variable
from utils import utils
from utils import plot
def generate_plots(exp, epoch):
plots = dict();
if exp.flags.factorized_representation:
# mnist to mnist: swapping content and style intra modal
swapping_figs = generate_swapping_plot(exp, epoch)
plots['swapping'] = swapping_figs;
for k in range(len(exp.modalities.keys())):
cond_k = generate_conditional_fig_M(exp, epoch, k+1)
plots['cond_gen_' + str(k+1).zfill(2)] = cond_k;
plots['random'] = generate_random_samples_plots(exp, epoch);
return plots;
def generate_random_samples_plots(exp, epoch):
model = exp.mm_vae;
mods = exp.modalities;
num_samples = 100;
random_samples = model.generate(num_samples)
random_plots = dict();
for k, m_key_in in enumerate(mods.keys()):
mod = mods[m_key_in];
samples_mod = random_samples[m_key_in];
rec = torch.zeros(exp.plot_img_size,
dtype=torch.float32).repeat(num_samples,1,1,1);
for l in range(0, num_samples):
rand_plot = mod.plot_data(samples_mod[l]);
rec[l, :, :, :] = rand_plot;
random_plots[m_key_in] = rec;
for k, m_key in enumerate(mods.keys()):
fn = os.path.join(exp.flags.dir_random_samples, 'random_epoch_' +
str(epoch).zfill(4) + '_' + m_key + '.png');
mod_plot = random_plots[m_key];
p = plot.create_fig(fn, mod_plot, 10, save_figure=exp.flags.save_figure);
random_plots[m_key] = p;
return random_plots;
def generate_swapping_plot(exp, epoch):
model = exp.mm_vae;
mods = exp.modalities;
samples = exp.test_samples;
swap_plots = dict();
for k, m_key_in in enumerate(mods.keys()):
mod_in = mods[m_key_in];
for l, m_key_out in enumerate(mods.keys()):
mod_out = mods[m_key_out];
rec = torch.zeros(exp.plot_img_size,
dtype=torch.float32).repeat(121,1,1,1);
rec = rec.to(exp.flags.device);
for i in range(len(samples)):
c_sample_in = mod_in.plot_data(samples[i][mod_in.name]);
s_sample_out = mod_out.plot_data(samples[i][mod_out.name]);
rec[i+1, :, :, :] = c_sample_in;
rec[(i + 1) * 11, :, :, :] = s_sample_out;
# style transfer
for i in range(len(samples)):
for j in range(len(samples)):
i_batch_s = {mod_out.name: samples[i][mod_out.name].unsqueeze(0)}
i_batch_c = {mod_in.name: samples[i][mod_in.name].unsqueeze(0)}
l_style = model.inference(i_batch_s,
num_samples=1)
l_content = model.inference(i_batch_c,
num_samples=1)
l_s_mod = l_style['modalities'][mod_out.name + '_style'];
l_c_mod = l_content['modalities'][mod_in.name];
s_emb = model.reparameterize(l_s_mod[0], l_s_mod[1]);
c_emb = model.reparameterize(l_c_mod[0], l_c_mod[1]);
style_emb = {mod_out.name: s_emb}
emb_swap = {'content': c_emb, 'style': style_emb};
swap_sample = model.generate_from_latents(emb_swap);
swap_out = mod_out.plot_data(swap_sample[mod_out.name].squeeze(0));
rec[(i+1) * 11 + (j+1), :, :, :] = swap_out;
fn_comb = (mod_in.name + '_to_' + mod_out.name + '_epoch_'
+ str(epoch).zfill(4) + '.png');
fn = os.path.join(exp.flags.dir_swapping, fn_comb);
swap_plot = plot.create_fig(fn, rec, 11, save_figure=exp.flags.save_figure);
swap_plots[mod_in.name + '_' + mod_out.name] = swap_plot;
return swap_plots;
def generate_conditional_fig_M(exp, epoch, M):
model = exp.mm_vae;
mods = exp.modalities;
samples = exp.test_samples;
subsets = exp.subsets;
# get style from random sampling
random_styles = model.get_random_styles(10);
cond_plots = dict();
for k, s_key in enumerate(subsets.keys()):
subset = subsets[s_key];
num_mod_s = len(subset);
if num_mod_s == M:
s_in = subset;
for l, m_key_out in enumerate(mods.keys()):
mod_out = mods[m_key_out];
rec = torch.zeros(exp.plot_img_size,
dtype=torch.float32).repeat(100 + M*10,1,1,1);
for m, sample in enumerate(samples):
for n, mod_in in enumerate(s_in):
c_in = mod_in.plot_data(sample[mod_in.name]);
rec[m + n*10, :, :, :] = c_in;
cond_plots[s_key + '__' + mod_out.name] = rec;
# style transfer
for i in range(len(samples)):
for j in range(len(samples)):
i_batch = dict();
for o, mod in enumerate(s_in):
i_batch[mod.name] = samples[j][mod.name].unsqueeze(0);
latents = model.inference(i_batch, num_samples=1)
c_in = latents['subsets'][s_key];
c_rep = model.reparameterize(mu=c_in[0], logvar=c_in[1]);
style = dict();
for l, m_key_out in enumerate(mods.keys()):
mod_out = mods[m_key_out];
if exp.flags.factorized_representation:
style[mod_out.name] = random_styles[mod_out.name][i].unsqueeze(0);
else:
style[mod_out.name] = None;
cond_mod_in = {'content': c_rep, 'style': style};
cond_gen_samples = model.generate_from_latents(cond_mod_in);
for l, m_key_out in enumerate(mods.keys()):
mod_out = mods[m_key_out];
rec = cond_plots[s_key + '__' + mod_out.name];
squeezed = cond_gen_samples[mod_out.name].squeeze(0);
p_out = mod_out.plot_data(squeezed);
rec[(i+M) * 10 + j, :, :, :] = p_out;
cond_plots[s_key + '__' + mod_out.name] = rec;
for k, s_key_in in enumerate(subsets.keys()):
subset = subsets[s_key_in];
if len(subset) == M:
s_in = subset;
for l, m_key_out in enumerate(mods.keys()):
mod_out = mods[m_key_out];
rec = cond_plots[s_key_in + '__' + mod_out.name];
fn_comb = (s_key_in + '_to_' + mod_out.name + '_epoch_' +
str(epoch).zfill(4) + '.png')
fn_out = os.path.join(exp.flags.dir_cond_gen, fn_comb);
plot_out = plot.create_fig(fn_out, rec, 10, save_figure=exp.flags.save_figure);
cond_plots[s_key_in + '__' + mod_out.name] = plot_out;
return cond_plots;