-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_and_render.py
200 lines (165 loc) · 8.1 KB
/
train_and_render.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
import os
import wandb
from gaussian_splatting.train import *
import torch
import numpy as np
from torchvision import transforms
from sklearn.metrics import roc_auc_score, precision_recall_curve, roc_curve
from scipy.ndimage import gaussian_filter
# needed for PAD code
from easydict import EasyDict
import yaml
classnames = ["01Gorilla", "02Unicorn", "03Mallard", "04Turtle", "05Whale", "06Bird", "07Owl", "08Sabertooth",
"09Swan", "10Sheep", "11Pig", "12Zalika", "13Pheonix", "14Elephant", "15Parrot", "16Cat", "17Scorpion",
"18Obesobeso", "19Bear", "20Puppy"]
pre_parser = ArgumentParser(description="Parameters of the LEGO training run")
pre_parser.add_argument("-k", metavar="K", type=int, help="number of pose estimation steps", default=175)
pre_parser.add_argument("-c", "-classname", metavar="c", type=str, help="current class to run experiments on",
default="01Gorilla")
pre_parser.add_argument("-wandb_config", metavar="WC", type=str, help="the wandb config to use", default="None")
pre_parser.add_argument("-p", "-prefix", metavar="pf", type=str, help="prefix for the wandb run name", default="to_delete")
pre_parser.add_argument("-seed", type=int, help="seed for random behavior", default=0)
pre_parser.add_argument("-gauss_iters", type=int, help="number of training iterations for 3DGS", default=30000)
pre_parser.add_argument("-wandb", type=int, help="whether we track with wandb", default=0)
pre_parser.add_argument("-train", type=int, help="whether we train or look for a saved model", default=1)
pre_parser.add_argument("-v", type=int, help="verbosity", default=0)
pre_parser.add_argument("-data_path", type=str, help="path pointing towards the usable data set", default="MAD-Sim_3dgs/")
lego_args = pre_parser.parse_args()
data_base_dir = lego_args.data_path
config = {
"k" : lego_args.k,
"classname" : lego_args.c,
"seed" : lego_args.seed,
"3dgs_iters" : lego_args.gauss_iters,
"prefix" : lego_args.p,
"wandb" : lego_args.wandb,
"train" : lego_args.train,
"data_dir" : data_base_dir,
"verbose" : lego_args.v != 0
}
projectname = config["prefix"]
if config["wandb"] != 0:
run = wandb.init(project=projectname, config=config, name=f"{config['prefix']}_{config['classname']}")
data_path = os.path.join(data_base_dir, config["classname"])
result_dir = os.path.join(data_base_dir, f"results_{config['prefix']}_{config['seed']}", config["classname"])
print("saving model to: ", result_dir)
os.makedirs(result_dir, exist_ok=True)
if config["train"] != 0:
start = torch.cuda.Event(enable_timing=True)
end = torch.cuda.Event(enable_timing=True)
start.record()
# Set up command line argument parser
training_args = ["-w", "--eval", "-s", data_path, "-m", result_dir, "--iterations", str(config["3dgs_iters"]), "--sh_degree", "0"]
print("training args: ", training_args)
parser = ArgumentParser(description="3DGS Training script parameters")
lp = ModelParams(parser)
op = OptimizationParams(parser)
pp = PipelineParams(parser)
parser.add_argument('--ip', type=str, default="127.0.0.1")
parser.add_argument('--port', type=int, default=6009)
parser.add_argument('--debug_from', type=int, default=-1)
parser.add_argument('--detect_anomaly', action='store_true', default=False)
parser.add_argument("--test_iterations", nargs="+", type=int, default=[7_000, config["3dgs_iters"]])
parser.add_argument("--save_iterations", nargs="+", type=int, default=[config["3dgs_iters"]])
parser.add_argument("--quiet", action="store_true")
parser.add_argument("--checkpoint_iterations", nargs="+", type=int, default=[])
parser.add_argument("--start_checkpoint", type=str, default = None)
args = parser.parse_args(training_args)
args.save_iterations.append(args.iterations)
print("Optimizing " + args.model_path)
# Initialize system state (RNG)
safe_state(args.quiet, config["seed"])
torch.autograd.set_detect_anomaly(args.detect_anomaly)
training(lp.extract(args), op.extract(args), pp.extract(args), args.test_iterations, args.save_iterations,
args.checkpoint_iterations, args.start_checkpoint, args.debug_from)
end.record()
torch.cuda.synchronize()
train_time_millis = start.elapsed_time(end)
if config["wandb"] != 0:
wandb.log({
"train_seconds" : train_time_millis / 1000
})
else:
if config["wandb"] != 0:
wandb.log({
"train_seconds" : 0
})
print("skipping training!")
from pose_estimation import main_pose_estimation
from utils_pose_est import ModelHelper, update_config
from aupro import calculate_au_pro_au_roc
test_images, reference_images, all_labels, gt_masks, times = main_pose_estimation(cur_class=config["classname"],
model_dir_location=result_dir,
k=config["k"], verbose=config["verbose"],
data_dir=None)
if config["wandb"] != 0:
my_data = [[i, times[i]] for i in range(len(times))]
columns = ["index", "time_millis"]
cur_table = wandb.Table(data=my_data, columns=columns)
wandb.log({"time_millis": cur_table})
with open("PAD_utils/config_effnet.yaml") as f:
mad_config = EasyDict(yaml.load(f, Loader=yaml.FullLoader))
mad_config = update_config(mad_config)
model = ModelHelper(mad_config.net)
model.eval()
model.cuda()
# evaluation Code taken from PAD/MAD data set paper at https://github.com/EricLee0224/PAD
criterion = torch.nn.MSELoss(reduction='none')
tf_img = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
tf_mask = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(224, interpolation=transforms.InterpolationMode.NEAREST),
])
test_imgs = list()
score_map_list=list()
scores=list()
pred_list=list()
recon_imgs=list()
with torch.no_grad():
for i in range(len(test_images)):
ref=tf_img(reference_images[i]).unsqueeze(0).cuda()
rgb=tf_img(test_images[i]).unsqueeze(0).cuda()
ref_feature=model(ref)
rgb_feature=model(rgb)
score = criterion(ref, rgb).sum(1, keepdim=True)
for i in range(len(ref_feature)):
s_act = ref_feature[i]
mse_loss = criterion(s_act, rgb_feature[i]).sum(1, keepdim=True)
score += torch.nn.functional.interpolate(mse_loss, size=224, mode='bilinear', align_corners=False)
score = score.squeeze(1).cpu().numpy()
for i in range(score.shape[0]):
score[i] = gaussian_filter(score[i], sigma=4)
recon_imgs.extend(rgb.cpu().numpy())
test_imgs.extend(ref.cpu().numpy())
scores.append(score)
scores = np.asarray(scores).squeeze()
max_anomaly_score = scores.max()
min_anomaly_score = scores.min()
scores = (scores - min_anomaly_score) / (max_anomaly_score - min_anomaly_score)
gt_mask = np.concatenate([np.asarray(tf_mask(a))[None,...] for a in gt_masks], axis=0)
precision, recall, thresholds = precision_recall_curve(gt_mask.flatten(), scores.flatten())
a = 2 * precision * recall
b = precision + recall
f1 = np.divide(a, b, out=np.zeros_like(a), where=b != 0)
threshold = thresholds[np.argmax(f1)]
fpr, tpr, _ = roc_curve(gt_mask.flatten(), scores.flatten())
per_pixel_rocauc = roc_auc_score(gt_mask.flatten(), scores.flatten())
print('pixel ROCAUC: %.3f' % (per_pixel_rocauc))
au_pro, au_roc, pro_curve, roc_curve = calculate_au_pro_au_roc(gt_mask, scores)
print(f"aupro: {au_pro}. and other au_roc: {au_roc}")
img_scores = scores.reshape(scores.shape[0], -1).max(axis=1)
gt_list_isano = np.asarray(all_labels) != 0
img_roc_auc = roc_auc_score(gt_list_isano, img_scores)
print('image ROCAUC: %.3f' % (img_roc_auc))
if config["wandb"] != 0:
wandb.log({
"avg_time" : np.mean(times),
"pixel_roc" : per_pixel_rocauc,
"image_roc" : img_roc_auc,
"aupro" : au_pro
})