-
Notifications
You must be signed in to change notification settings - Fork 1
/
DAM_gen_one_sample.py
186 lines (140 loc) · 6.6 KB
/
DAM_gen_one_sample.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
#!/usr/bin/env python
# coding: utf-8
import argparse
import numpy as np
import time
import utils
#get_ipython().run_line_magic('matplotlib', 'inline')
import torch
import struct
from utils.ModelNetDataLoader import ModelNetDataLoader
from write_ply import write_pointcloud
from torch.autograd import Variable
import os
import importlib
import torch.optim as optim
import sys
import random
from models.vae_flow import *
from write_ply import write_pointcloud
from models.common import *
parser = argparse.ArgumentParser()
# =============================================================================
# parser.add_argument('--ckpt', type=str, default='./logs_gen/PDT_Shapenet2023_06_29__12_05_35/ckpt_0.000000_1020000.pt')
# =============================================================================
parser.add_argument('--ckpt', type=str, default='./logs_gen/Label_PDT_DS2502023_05_25__17_34_55/ckpt_0.000000_1020000.pt')
parser.add_argument('--device', type=str, default='cuda' if torch.cuda.is_available() else 'cpu')
# Datasets and loaders
parser.add_argument('--dataset', type=str, default='ModelNet40', choices=['ModelNet40', 'ShapeNet'])
# Sampling
parser.add_argument('--sample_num_points', type=int, default=1024)
parser.add_argument('--normalize', type=str, default='shape_bbox', choices=[None, 'shape_unit', 'shape_bbox'])
parser.add_argument('--num_points', type=int, default=1024)
parser.add_argument('--normal', default=False)
args = parser.parse_args()
if args.device == 'cpu':
ckpt = torch.load(args.ckpt, map_location=torch.device('cpu'))
else:
ckpt = torch.load(args.ckpt)
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
ROOT_DIR = BASE_DIR
sys.path.append(os.path.join(ROOT_DIR, 'models/classifier'))
# =============================================================================
# def add_sparse_Gauss_noise(vector,noise_mean,noise_var,noise_rate=0.05, n_weight = 0):
# v_shape = vector.size()
# #G_noise = torch.empty(v_shape,device = device).normal_(mean=noise_mean,std=noise_var)
# G_noise = torch.empty(v_shape).normal_(mean=noise_mean,std=noise_var)
# noise_mask = torch.empty(v_shape).uniform_() > 0.95
# G_noise = torch.mul(G_noise,noise_mask))
# res = vector + torch.Tensor(1)#(n_weight * G_noise.to(device))
# return res
# =============================================================================
shape_names = []
SHAPE_NAME_FILE = 'data/shape_names.txt'
with open(SHAPE_NAME_FILE, "r") as f:
for tmp in f.readlines():
tmp = tmp.strip('\n')
shape_names.append(tmp)
batch_size = 1
output_folder = "output/" # folder path to save the results
save_results = True # save the results to output_folder
latent_size = 128 # bottleneck size of the Autoencoder model
if args.dataset == 'ModelNet40':
num_class = 40
elif args.dataset == 'ShapeNet':
num_class = 55
tar_label = 0
avg_initialization = True
maximize_logstfmx = False
lr = 1e-3
# =============================================================================
# from pytorch3d.loss import chamfer_distance # chamfer distance for calculating point cloud distance
# =============================================================================
if args.dataset == 'ModelNet40':
model_name = os.listdir('log_classifier/classification/pointnet_cls_msg'+'/logs')[0].split('.')[0]
elif args.dataset == 'ShapeNet':
model_name = os.listdir('log_classifier/classification_shapenet/pointnet_cls_shapenet'+'/logs')[0].split('.')[0]
MODEL = importlib.import_module(model_name)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# =============================================================================
# ckpt['args'].device = 'cpu'
# =============================================================================
generator = PDT(ckpt['args']).to(args.device)
generator.load_state_dict(ckpt['state_dict'])
print("Generator: ", generator, "\n\n")
MODEL = importlib.import_module('pointnet_cls')
classifier = MODEL.get_model(num_class,normal_channel=False)
classifier = classifier.eval()
if args.dataset == 'ModelNet40':
cls_path = 'log_classifier/classification/pointnet_cls_msg/checkpoints/best_model.pth'
elif args.dataset == 'ShapeNet':
cls_path = 'log_classifier/classification_shapenet/pointnet_cls_shapenet/checkpoints/best_model.pth'
checkpoint_cls = torch.load(cls_path, map_location=torch.device(args.device))
classifier.load_state_dict(checkpoint_cls['model_state_dict'])
print("Loading model: ", cls_path)
print("Classifier: ", classifier, "\n\n")
for param in classifier.parameters():
param.requires_grad = False
MODEL_guide = importlib.import_module('pointnet_cls_t')
classifier_noised_t = MODEL_guide.get_model(num_class,normal_channel=False, channel=11)
classifier_noised_t = classifier_noised_t.eval()
if args.dataset == 'ModelNet40':
noised_path = 'log_classifier/classification/pointnet_cls_msg/checkpoints/best_model_noised_t.pth'
elif args.dataset == 'ShapeNet':
noised_path = 'log_classifier/classification_shapenet/pointnet_cls_shapenet/checkpoints/best_model_noised_t.pth'
checkpoint_noised_t = torch.load(noised_path, map_location=torch.device(args.device))
classifier_noised_t.load_state_dict(checkpoint_noised_t['model_state_dict'])
print("Loading model: ", noised_path)
print("Classifier: ", classifier_noised_t, "\n\n")
for param in classifier_noised_t.parameters():
param.requires_grad = False
generator = generator.to(args.device)
classifier = classifier.to(args.device)
classifier_noised_t = classifier_noised_t.to(args.device)
for param_g in generator.parameters():
param_g.requires_grad = False
num_samples = 20
tar_cls = 33
AM_lr = 1e-4
label = torch.zeros([num_samples,1])
label = label + tar_cls
label = label.long().to(args.device)
x = torch.randn([num_samples,1024,3]).to(args.device)
z_mu, z_sigma = generator.encoding(x)
z = reparameterize_gaussian(mean=z_mu, logvar=z_sigma)
EXP = generator.AM_sample(z, label, AM_lr, args.sample_num_points, classifier=classifier, cls_noise_guide=classifier_noised_t, flexibility=ckpt['args'].flexibility)
if args.device == 'cpu':
EXP_npy = EXP.detach().numpy()
else:
EXP_npy = EXP.detach().cpu().numpy()
for i in range(num_samples):
if args.dataset == 'ModelNet40':
write_pointcloud("./visu_EXP/%s_%d.ply" % (shape_names[tar_cls], i), EXP_npy[i])
elif args.dataset == 'ShapeNet':
write_pointcloud("./visu_EXP/%d_%d.ply" % (tar_cls, i), EXP_npy[i])
EXP = EXP.transpose(1,2)
pred, bf_sfx, _ = classifier(EXP)
pred_cls = torch.argmax(pred, dim=1).detach().cpu().numpy()
bf_sfx = bf_sfx.detach().cpu().numpy()
for i in range(num_samples):
print(bf_sfx[i][tar_cls], shape_names[pred_cls[i]])