-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_strands_optimization.py
202 lines (152 loc) · 8.68 KB
/
run_strands_optimization.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
import errno
import os
import sys
from collections import namedtuple
import random
sys.path.append('..')
import argparse
import logging
import math
import time
from datetime import datetime
from pathlib import Path
from shutil import copyfile
import cv2 as cv
import numpy as np
import torch
import torch.nn.functional as F
import trimesh
from src.models.dataset import Dataset, MonocularDataset
from pyhocon import ConfigFactory
from torch.utils.tensorboard import SummaryWriter
from tqdm import tqdm
from src.hair_networks.sdf import HairSDFNetwork
from src.strands_trainer import StrandsTrainer
import yaml
from copy import deepcopy
from src.utils.util import set_seed, scale_mat
import warnings
warnings.filterwarnings("ignore")
class Runner:
def __init__(self, conf_path, case='CASE_NAME', scene_type='DATASET_TYPE', checkpoint_name=None, hair_conf_path=None, exp_name=None):
self.device = torch.device('cuda')
# Configuration of geometry
self.conf_path = conf_path
with open(conf_path, 'r') as f:
replaced_conf = str(yaml.load(f, Loader=yaml.Loader)).replace('CASE_NAME', case)
self.conf = yaml.load(replaced_conf, Loader=yaml.Loader)
# Configuration of hair strands
self.hair_conf_path = hair_conf_path
with open(hair_conf_path, 'r') as f:
replaced_conf = str(yaml.load(f, Loader=yaml.Loader)).replace('CASE_NAME', case)
replaced_conf = replaced_conf.replace('DATASET_TYPE', scene_type)
self.hair_conf = yaml.load(replaced_conf, Loader=yaml.Loader)
train_conf = self.conf['train']
self.end_iter = train_conf['end_iter']
self.report_freq = train_conf['report_freq']
self.batch_size = train_conf['batch_size']
if exp_name is not None:
date, time = str(datetime.today()).split('.')[0].split(' ')
exps_dir = Path('./exps_second_stage') / exp_name / case / Path(conf_path).stem
cur_dir = date + '_' + time
self.base_exp_dir = exps_dir / cur_dir
else:
self.base_exp_dir = self.conf['general']['base_exp_dir']
self.img_size = self.hair_conf['render']['image_size']
os.makedirs(self.base_exp_dir, exist_ok=True)
os.makedirs(os.path.join(self.base_exp_dir, 'meshes'), exist_ok=True)
os.makedirs(os.path.join(self.base_exp_dir, 'hair_primitives'), exist_ok=True)
if scene_type == 'h3ds':
self.dataset = Dataset(self.conf['dataset'])
else:
self.dataset = MonocularDataset(self.conf['dataset'])
self.iter_step = 0
self.writer = None
set_seed(42)
self.hair_primitives_trainer = StrandsTrainer(self.hair_conf, run_model= lambda model, x: model(x, calc_orient=True), device=self.device, save_dir=self.base_exp_dir)
self.hair_network = HairSDFNetwork(**self.conf['model']['hair_sdf_network']).to(self.device)
# Upload volumetric geometry and surface orientation fields
if train_conf['pretrain_path']:
print('Upload sdf hair geometry and orientation field!')
checkpoint = torch.load(train_conf['pretrain_path'], map_location=self.device)
self.hair_network.load_state_dict(checkpoint['hair_network'])
# Upload strand-based geometry
if train_conf['pretrain_strands_path']:
print('Upload strands!')
self.hair_primitives_trainer.load_weights(train_conf['pretrain_hair_path'])
# Backup codes and configs for debug
self.file_backup()
def train(self):
res_step = self.end_iter - self.iter_step
self.writer = SummaryWriter(log_dir=os.path.join(self.base_exp_dir, 'logs'))
image_perm = self.get_image_perm()
losses = {}
bound_min = torch.tensor(self.dataset.object_bbox_min, dtype=torch.float32)
bound_max = torch.tensor(self.dataset.object_bbox_max, dtype=torch.float32)
for iter_i in tqdm(range(res_step)):
cur_img = image_perm[self.iter_step % len(image_perm)]
_, cam_intr, cam_pose = self.dataset.gen_random_rays_at(cur_img, self.batch_size)
cam_extr = torch.linalg.inv(cam_pose.detach())
orig_data_size = self.dataset.images[cur_img].shape[1] # H shape, suppose square images
raster_dict = {}
raster_dict['iter'] = self.iter_step
raster_dict['cam_extr'] = cam_extr
if orig_data_size is None:
raster_dict['cam_intr'] = cam_intr
raster_dict['gt_silh'] = self.dataset.hair_masks[cur_img].permute(2, 0, 1).cuda()
raster_dict['gt_rgb'] = self.dataset.images[cur_img].permute(2, 0, 1).cuda()
else:
# need to change cameras intrinsic as we render in resolution 512x512
scale_factor = orig_data_size / self.img_size
raster_dict['cam_intr'] = scale_mat(deepcopy(cam_intr), scale_factor)
raster_dict['gt_silh'] = F.interpolate(self.dataset.hair_masks[cur_img].permute(2, 0, 1)[None], size=self.img_size, mode='bilinear')[0].cuda()
raster_dict['gt_rgb'] = F.interpolate(self.dataset.images[cur_img].permute(2, 0, 1)[None], size=self.img_size, mode='bilinear')[0].cuda()
raster_dict['visual_gt_orients'] = self.dataset.orient_at(cur_img, resolution_level=1)
losses.update({
'hair_' + str(key): val for key, val in self.hair_primitives_trainer.train_step(model=self.hair_network, it=self.iter_step, raster_dict=raster_dict).items()
})
self.iter_step += 1
if losses is not None:
for k, v in losses.items():
self.writer.add_scalar(f'Loss/{k}', v, self.iter_step)
if self.iter_step % self.report_freq == 0:
self.save_strands_pointcloud()
if self.iter_step % len(image_perm) == 0:
image_perm = self.get_image_perm()
if self.iter_step % self.report_freq == 0:
self.hair_primitives_trainer.save_weights(os.path.join(self.base_exp_dir, 'hair_primitives', 'ckpt_{:0>6d}.pth'.format(self.iter_step)))
def save_strands_pointcloud(self):
if self.hair_primitives_trainer:
strands_origins = self.hair_primitives_trainer.strands_origins.reshape(-1, 100, 3)
cols = torch.cat((torch.rand(strands_origins.shape[0], 3).unsqueeze(1).repeat(1, 100, 1), torch.ones(strands_origins.shape[0], 100, 1)), dim=-1).reshape(-1, 4).cpu()
trimesh.PointCloud(strands_origins.reshape(-1, 3).detach().cpu(), colors=cols).export(os.path.join(self.base_exp_dir, 'meshes', '{:0>8d}_strands_points.ply'.format(self.iter_step)))
def get_image_perm(self):
return torch.randperm(self.dataset.n_images)
def file_backup(self):
dir_lis = self.conf['general']['recording']
os.makedirs(os.path.join(self.base_exp_dir, 'recording'), exist_ok=True)
for dir_name in dir_lis:
cur_dir = os.path.join(self.base_exp_dir, 'recording', dir_name)
os.makedirs(cur_dir, exist_ok=True)
files = os.listdir(dir_name)
for f_name in files:
if f_name[-3:] == '.py':
copyfile(os.path.join(dir_name, f_name), os.path.join(cur_dir, f_name))
copyfile(self.conf_path, os.path.join(self.base_exp_dir, 'recording', 'config.yaml'))
copyfile(self.hair_conf_path, os.path.join(self.base_exp_dir, 'recording', 'hair_config.yaml'))
if __name__ == '__main__':
print('Hello Wooden')
FORMAT = "[%(filename)s:%(lineno)s - %(funcName)20s() ] %(message)s"
logging.basicConfig(level=logging.DEBUG, format=FORMAT)
parser = argparse.ArgumentParser()
parser.add_argument('--conf', type=str, default='./confs/base.conf')
parser.add_argument('--gpu', type=int, default=0)
parser.add_argument('--case', type=str, default='')
parser.add_argument('--scene_type', type=str, default='')
parser.add_argument('--hair_conf', type=str, default=None, help='Use hair primitives config')
parser.add_argument('--checkpoint', type=str, default=None, help='Checkpoint to continue training')
parser.add_argument('--exp_name', type=str, default=None)
args = parser.parse_args()
torch.cuda.set_device(args.gpu)
runner = Runner(args.conf, args.case, args.scene_type, hair_conf_path=args.hair_conf, checkpoint_name=args.checkpoint, exp_name=args.exp_name)
runner.train()