-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathvalscene_inference.py
181 lines (163 loc) · 7.68 KB
/
valscene_inference.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
import argparse
import os
import time
import datetime
import torch
import trimesh
import numpy as np
import torch.nn.functional as F
# from torch.utils.data import DataLoader
from torch_geometric.loader import DataLoader # for depth Batch
from loguru import logger
from utils import tensor2float, make_nograd_func, SaveScene
from datasets import transforms
from configs import cfg, update_config
from eval.evaluation import Renderer
from datasets.scannet import ScanNetDataset
from src.cdrnet import CrossDimensionalRefmntNet
def args():
parser = argparse.ArgumentParser(description='Testing for CDRNet model')
parser.add_argument('--cfg',
help='experiment configuration file',
default='configs/inference_scene0249.yaml',
type=str)
parser.add_argument('--save_path', type=str, default=None)
arguments = parser.parse_args()
return arguments
args_object = args()
update_config(cfg, args_object)
# for determinism
torch.manual_seed(cfg.SEED)
torch.cuda.manual_seed(cfg.SEED)
# create logger
current_time_str = str(datetime.datetime.now().strftime('%Y%m%d_%H%M%S'))
if not cfg.DEBUG:
logdir = cfg.LOGDIR
if not os.path.isdir(cfg.LOGDIR):
os.makedirs(cfg.LOGDIR)
elif cfg.LOGDIR is None:
os.makedirs(f'inference_{current_time_str}', exist_ok=False)
logdir = f'inference_{current_time_str}'
logfile_path = os.path.join(logdir, f'{current_time_str}_{cfg.MODE}.log')
logger.add(logfile_path, format='{time} {level} {message}', level="INFO", rotation="12:00")
# data augmentation
n_views = cfg.TEST.N_VIEWS
random_rotation = False
random_translation = False
paddingXY = 0
paddingZ = 0
transform = []
transform += [transforms.ResizeImageAndLabel((640, 480)),
transforms.ToTensor(),
transforms.RandomTransformSpace(
cfg.MODEL.N_VOX,
cfg.MODEL.VOXEL_SIZE,
random_rotation,
random_translation,
paddingXY,
paddingZ),
transforms.IntrinsicsPoseToProjection(n_views, 4)]
transforms = transforms.Compose(transform)
# single scene dataset
single_scene_scannet = ScanNetDataset(datapath=cfg.TEST.PATH,
fragpath='data/meta_data/scene0249_00/fragments.pkl',
pklpath=cfg.PKL_PATH,
mode='val',
transforms=transforms,
nviews=cfg.TEST.N_VIEWS,
n_scales=len(cfg.MODEL.THRESHOLDS) - 1,
datasplit='val',
depth_prediction=cfg.MODEL.DEPTH_PREDICTION,
semseg_matching=cfg.MODEL.CDR.SEMSEG_REFMNT,
nyu40_20_conversion=cfg.MODEL.CDR.SEMSEG_CLASS_3D == 20
)
inference_loader = DataLoader(single_scene_scannet,
batch_size=cfg.BATCH_SIZE,
num_workers=cfg.TEST.N_WORKERS,
shuffle=False,
pin_memory=True,
drop_last=False)
model = CrossDimensionalRefmntNet(cfg).cuda().eval()
model = torch.nn.DataParallel(model, device_ids=[0])
model_49 = CrossDimensionalRefmntNet(cfg).cuda().eval()
model_49 = torch.nn.DataParallel(model_49, device_ids=[0])
save_mesh = SaveScene(cfg)
logger.info(model)
logger.info('Cfg is as below: {}\n'.format(cfg))
@make_nograd_func
def inference_sample(sample, save_scene=False, duration=None):
model.eval()
start_time = time.time()
outputs, vis_metrics = model(sample, save_scene)
duration += time.time() - start_time
loss = vis_metrics['total_loss']
tsdf_loss = vis_metrics['total_tsdf_occ_loss']
if 'total_3d_semseg_loss' in vis_metrics.keys():
semseg_loss = vis_metrics['total_3d_semseg_loss']
if 'total_3d_semseg_loss' in vis_metrics.keys():
return tensor2float(loss), outputs, tensor2float(tsdf_loss), tensor2float(semseg_loss), duration
def inference(loadckpt):
logger.info('resuming ' + str(loadckpt))
state_dict = torch.load(loadckpt)
model.load_state_dict(state_dict['model'], strict=False)
# state_dict_49 = torch.load('model_000049.ckpt')
# model_49.load_state_dict(state_dict_49, strict=False)
epoch_idx = state_dict['epoch']
logger.info('Conducting inference, with model epoch {}'.format(epoch_idx))
batch_len = len(inference_loader)
tsdf_loss_list = []
semseg_loss_list = []
total_loss_list = []
duration_time = 0
if cfg.MODEL.VIS_DEPTH: # to create renderer based on the predicted mesh, vis depth needed to be one secondly
mesh_to_render = 'logs_viz_one_scene_valset/val/scene_scannet_fusion_eval_49_with_depth_jan17/scene0249_00.ply'
mesh_to_render = trimesh.load(mesh_to_render, process=False)
renderer = Renderer()
mesh_opengl = renderer.mesh_opengl(mesh_to_render)
with torch.no_grad():
for batch_idx, sample in enumerate(inference_loader):
save_scene = cfg.SAVE_SCENE_MESH and batch_idx == batch_len - 1
if cfg.MODEL.VIS_DEPTH: # to create the depth_from_mesh tensor
h, w = 480, 640
cam_intr = sample['intrinsics']
cam_pose = sample['poses']
depth_from_mesh_list = []
for i in range(9):
_, depth_from_mesh = renderer(h, w,
cam_intr[0, i], cam_pose[0, i],
mesh_opengl)
depth_from_mesh_list.append(depth_from_mesh)
depth_from_mesh = torch.FloatTensor(np.stack(depth_from_mesh_list))
depth_from_mesh = F.interpolate(depth_from_mesh.unsqueeze(0), scale_factor=1 / 4,
mode='nearest').squeeze(0)
sample['depth_from_mesh'] = depth_from_mesh
start_time = time.time()
loss, outputs, tsdf_loss, semseg_loss, duration_time = \
inference_sample(sample, save_scene=save_scene, duration=duration_time)
logger.info('{}, epoch {}, iter {}/{}, tsdf loss = {:.3f}, semseg loss = {:.3f}, '
'time = {:.3f}'.format(sample['scene'][0], epoch_idx, batch_idx,
len(inference_loader), tsdf_loss, semseg_loss,
time.time() - start_time))
if os.path.isfile(loadckpt):
save_mesh(outputs, sample, epoch_idx)
tsdf_loss_list.append(tsdf_loss)
semseg_loss_list.append(semseg_loss)
total_loss_list.append(loss)
del tsdf_loss, semseg_loss, loss
loss_mean_epoch = sum(total_loss_list) / len(total_loss_list)
tsdf_mean_epoch = sum(tsdf_loss_list) / len(tsdf_loss_list)
semseg_mean_epoch = sum(semseg_loss_list) / len(semseg_loss_list)
return loss_mean_epoch, tsdf_mean_epoch, semseg_mean_epoch, duration_time
if __name__ == '__main__':
frag_len = len(inference_loader)
loss_mean, tsdf_mean, semseg_mean, duration = inference(loadckpt=cfg.LOADCKPT)
summary_text = f"""
Summary:
Total number of fragments: {frag_len}
Total tsdf loss: {tsdf_mean}
Total semseg loss: {semseg_mean}
Total loss: {loss_mean}
Average keyframes/sec: {1 / (duration / (frag_len * cfg.TEST.N_VIEWS))}
Total runtime: {duration}
"""
logger.info(summary_text)