-
Notifications
You must be signed in to change notification settings - Fork 56
/
Copy pathdemo_latent_space.py
185 lines (140 loc) · 6.74 KB
/
demo_latent_space.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
from util import device, ensure_directory
import scipy.interpolate
import numpy as np
from rendering import MeshRenderer
import torch
from tqdm import tqdm
import cv2
import random
import matplotlib.pyplot as plt
from sklearn.manifold import TSNE
from matplotlib.offsetbox import Bbox
from sklearn.cluster import KMeans
SAMPLE_COUNT = 30 # Number of distinct objects to generate and interpolate between
TRANSITION_FRAMES = 60
USE_VAE = False
SURFACE_LEVEL = 0.011
FRAMES = SAMPLE_COUNT * TRANSITION_FRAMES
progress = np.arange(FRAMES, dtype=float) / TRANSITION_FRAMES
if USE_VAE:
from model.autoencoder import Autoencoder, LATENT_CODE_SIZE
vae = Autoencoder()
vae.load()
vae.eval()
print("Calculating latent codes...")
from datasets import VoxelDataset
from torch.utils.data import DataLoader
dataset = VoxelDataset.glob('data/chairs/voxels_32/**.npy')
dataloader = DataLoader(dataset, batch_size=1000, num_workers=8)
latent_codes = torch.zeros((len(dataset), LATENT_CODE_SIZE))
with torch.no_grad():
position = 0
for batch in tqdm(dataloader):
latent_codes[position:position + batch.shape[0], :] = vae.encode(batch.to(device)).detach().cpu()
latent_codes = latent_codes.numpy()
else:
from model.sdf_net import SDFNet, LATENT_CODES_FILENAME
latent_codes = torch.load(LATENT_CODES_FILENAME).detach().cpu().numpy()
sdf_net = SDFNet()
sdf_net.load()
sdf_net.eval()
from shapenet_metadata import shapenet
raise NotImplementedError('A labels tensor needs to be supplied here.')
labels = None
print("Calculating embedding...")
tsne = TSNE(n_components=2)
latent_codes_embedded = tsne.fit_transform(latent_codes)
print("Calculating clusters...")
kmeans = KMeans(n_clusters=SAMPLE_COUNT)
indices = np.zeros(SAMPLE_COUNT, dtype=int)
kmeans_clusters = kmeans.fit_predict(latent_codes_embedded)
for i in range(SAMPLE_COUNT):
center = kmeans.cluster_centers_[i, :]
cluster_classes = labels[kmeans_clusters == i]
cluster_class = np.bincount(cluster_classes).argmax()
dist = np.linalg.norm(latent_codes_embedded - center[np.newaxis, :], axis=1)
dist[labels != cluster_class] = float('inf')
indices[i] = np.argmin(dist)
def try_find_shortest_roundtrip(indices):
best_order = indices
best_distance = None
for _ in range(5000):
candiate = best_order.copy()
a = random.randint(0, SAMPLE_COUNT-1)
b = random.randint(0, SAMPLE_COUNT-1)
candiate[a] = best_order[b]
candiate[b] = best_order[a]
dist = np.sum(np.linalg.norm(latent_codes_embedded[candiate, :] - latent_codes_embedded[np.roll(candiate, 1), :], axis=1)).item()
if best_distance is None or dist < best_distance:
best_distance = dist
best_order = candiate
return best_order, best_distance
def find_shortest_roundtrip(indices):
best_order, best_distance = try_find_shortest_roundtrip(indices)
for _ in tqdm(range(100)):
np.random.shuffle(indices)
order, distance = try_find_shortest_roundtrip(indices)
if distance < best_distance:
best_order = order
return best_order
print("Calculating trip...")
indices = find_shortest_roundtrip(indices)
indices = np.concatenate((indices, indices[0][np.newaxis]))
SIZE = latent_codes.shape[0]
stop_latent_codes = latent_codes[indices, :]
colors = np.zeros((labels.shape[0], 3))
for i in range(labels.shape[0]):
colors[i, :] = shapenet.get_color(labels[i])
spline = scipy.interpolate.CubicSpline(np.arange(SAMPLE_COUNT + 1), stop_latent_codes, axis=0, bc_type='periodic')
frame_latent_codes = spline(progress)
color_spline = scipy.interpolate.CubicSpline(np.arange(SAMPLE_COUNT + 1), colors[indices, :], axis=0, bc_type='periodic')
frame_colors = color_spline(progress)
frame_colors = np.clip(frame_colors, 0, 1)
frame_colors = np.zeros((progress.shape[0], 3))
for i in range(SAMPLE_COUNT):
frame_colors[i*TRANSITION_FRAMES:(i+1)*TRANSITION_FRAMES, :] = np.linspace(colors[indices[i]], colors[indices[i+1]], num=TRANSITION_FRAMES)
embedded_spline = scipy.interpolate.CubicSpline(np.arange(SAMPLE_COUNT + 1), latent_codes_embedded[indices, :], axis=0, bc_type='periodic')
frame_latent_codes_embedded = embedded_spline(progress)
frame_latent_codes_embedded[0, :] = frame_latent_codes_embedded[-1, :]
width, height = 40, 40
PLOT_FILE_NAME = 'tsne.png'
ensure_directory('images')
margin = 2
range_x = (latent_codes_embedded[:, 0].min() - margin, latent_codes_embedded[:, 0].max() + margin)
range_y = (latent_codes_embedded[:, 1].min() - margin, latent_codes_embedded[:, 1].max() + margin)
plt.ioff()
def create_plot(index, resolution=1080, filename=PLOT_FILE_NAME, dpi=100):
frame_color = frame_colors[index, :]
frame_color = (frame_color[0], frame_color[1], frame_color[2], 1.0)
size_inches = resolution / dpi
fig, ax = plt.subplots(1, figsize=(size_inches, size_inches), dpi=dpi)
ax.set_position([0, 0, 1, 1])
plt.axis('off')
ax.set_xlim(range_x)
ax.set_ylim(range_y)
ax.plot(frame_latent_codes_embedded[:, 0], frame_latent_codes_embedded[:, 1], c=(0.2, 0.2, 0.2, 1.0), zorder=1, linewidth=2)
ax.scatter(latent_codes_embedded[:, 0], latent_codes_embedded[:, 1], c=colors[:SIZE], s = 10, zorder=0)
ax.scatter(frame_latent_codes_embedded[index, 0], frame_latent_codes_embedded[index, 1], facecolors=frame_color, s = 200, linewidths=2, edgecolors=(0.1, 0.1, 0.1, 1.0), zorder=2)
ax.scatter(latent_codes_embedded[indices, 0], latent_codes_embedded[indices, 1], facecolors=colors[indices, :], s = 140, linewidths=1, edgecolors=(0.1, 0.1, 0.1, 1.0), zorder=3)
fig.savefig(filename, bbox_inches=Bbox([[0, 0], [size_inches, size_inches]]), dpi=dpi)
plt.close(fig)
frame_latent_codes = torch.tensor(frame_latent_codes, dtype=torch.float32, device=device)
print("Rendering...")
viewer = MeshRenderer(size=1080, start_thread=False)
def render_frame(frame_index):
viewer.model_color = frame_colors[frame_index, :]
with torch.no_grad():
if USE_VAE:
viewer.set_voxels(vae.decode(frame_latent_codes[frame_index, :]))
else:
viewer.set_mesh(sdf_net.get_mesh(frame_latent_codes[frame_index, :], voxel_resolution=128, sphere_only=True, level=SURFACE_LEVEL))
image_mesh = viewer.get_image(flip_red_blue=True)
create_plot(frame_index)
image_tsne = plt.imread(PLOT_FILE_NAME)[:, :, [2, 1, 0]] * 255
image = np.concatenate((image_mesh, image_tsne), axis=1)
cv2.imwrite("images/frame-{:05d}.png".format(frame_index), image)
for frame_index in tqdm(range(SAMPLE_COUNT * TRANSITION_FRAMES)):
render_frame(frame_index)
frame_index += 1
print("\n\nUse this command to create a video:\n")
print('ffmpeg -framerate 30 -i images/frame-%05d.png -c:v libx264 -profile:v high -crf 19 -pix_fmt yuv420p video.mp4')