-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathevaluate.py
177 lines (142 loc) · 8.21 KB
/
evaluate.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
import os
import argparse
from tqdm import tqdm
import warnings
import matplotlib.pyplot as plt
import torch
from mmengine.config import Config
from mmengine.runner import Runner
from mmseg.apis import init_model
from mmengine.logging.logger import MMLogger
from dood.models import *
from dood.datasets import *
from dood.utils.reference_features import get_reference_features, fwd_pass_mmseg
from dood.probes import get_feature_probe
from dood.utils.evaluation import ood_score_functions, upsample_scores, combine_scores_standardize, StreamingEval
from dood.utils.diffusion import get_diffusion_model, load_diffusion_checkpoint, get_diffusion_scores, get_diffusion_stats, get_denoising_steps_from_string
logger = MMLogger.get_instance('mmengine', log_level='INFO')
def parse_args():
parser = argparse.ArgumentParser(description='Evaluate a DOoD model for out-of-distribution detection.')
# setup
parser.add_argument('--model_config', required=True, type=str, help='Path to the config file')
parser.add_argument('--test_data_config', required=True, type=str, help='Path to the test data config file')
parser.add_argument('--segmentation_ckpt', required=True, type=str, help='Path to the segmentation model checkpoint file')
parser.add_argument('--diffusion_ckpt', required=True, type=str, help='Path to the diffusion model checkpoint file')
# reference data config: if not provided, it will looked for in the model config
parser.add_argument('--reference_data_config', required=False, type=str, help='Path to the reference data config file')
# evaluation
parser.add_argument('--num_reference_samples', type=int, default=25, help='Number of reference samples for normalization statistics.')
parser.add_argument('--parametric_score_fn', default="logsumexp")
parser.add_argument('--diffusion_score_fn', default="eps_cos", choices=["eps_mse", "eps_cos", "recon_mse"])
parser.add_argument('--num_diffusion_steps', default=25, type=int)
# save
parser.add_argument('--save_dir', default=None, type=str, help='Directory to save scoremaps')
args = parser.parse_args()
if args.save_dir is not None:
os.makedirs(args.save_dir, exist_ok=True)
logger.info(f"Saving scoremaps to {args.save_dir}")
return args
@torch.no_grad()
def get_reference_stats(model, diffusion_model, reference_dataset, feature_probe,
num_reference_samples,
parametric_score_fn,
diffusion_score_fn,
diffusion_steps):
reference_features, _, (param_mean_score, param_std_score) = get_reference_features(
model, reference_dataset, feature_probe,
num_samples=num_reference_samples,
ood_scoring_fn_for_scaling=ood_score_functions[parametric_score_fn],
reduce_zero_label=reference_dataset.reduce_zero_label,
seed=0)
mean_diff_score, std_diff_score = get_diffusion_stats(
reference_features,
diffusion_model,
diffusion_steps=diffusion_steps,
ood_eval_scores_type=diffusion_score_fn,
ft_key=list(reference_features.keys())[0])
return {"reference_features": reference_features,
"param_mean_std_scores": (param_mean_score, param_std_score),
"diff_mean_std_scores": (mean_diff_score, std_diff_score)}
@torch.no_grad()
def eval_ood(model, diffusion_model, test_dataset, feature_probe,
parametric_score_fn,
diffusion_score_fn,
diffusion_steps,
param_scores_mean_std=None,
diff_scores_mean_std=None,
save_dir=None, save_scoremaps=False,
dtype=torch.float16):
model.eval()
diffusion_model.eval()
diffusion_model.denoiser.to(dtype=dtype)
param_evaluator = StreamingEval(ood_id=test_dataset.ood_index, ignore_ids=255)
diff_evaluator = StreamingEval(ood_id=test_dataset.ood_index, ignore_ids=255)
evaluator = StreamingEval(ood_id=test_dataset.ood_index, ignore_ids=255)
progress = tqdm(total=len(test_dataset))
for sample_index, sample in enumerate(test_dataset):
if test_dataset.reduce_zero_label:
raise NotImplementedError
segm = sample['data_samples'].gt_sem_seg.data.squeeze(0)
logits = fwd_pass_mmseg(sample, model)[None,:]
param_scores = upsample_scores(ood_score_functions[parametric_score_fn](logits)[0], segm)
for k, test_ft in feature_probe.get_features().items():
h, w, c = test_ft.shape
# normalize before (so we can set a custom dtype), get scores
test_ft = diffusion_model.normalize(test_ft.view(h*w, c))
scores, scores_per_step = get_diffusion_scores(test_ft.view(h*w, c), diffusion_model, diffusion_steps, diffusion_score_fn, normalize=False, dtype=dtype)
diff_scores = upsample_scores(scores.view(h, w), segm)
param_evaluator.add(param_scores.cpu(), segm)
diff_evaluator.add(diff_scores.cpu(), segm)
comb_scores = combine_scores_standardize(param_scores, diff_scores, *param_scores_mean_std, *diff_scores_mean_std)
if args.save_dir is not None:
plt.imsave(os.path.join(args.save_dir, f"dood_scores_{sample_index:03d}.png"), comb_scores.squeeze().cpu().numpy(), cmap="turbo")
evaluator.add(comb_scores.cpu(), segm)
progress.update()
progress.close()
logger.info(f"Computing metrics...")
# logger.info("Parametric - AP: {}, FPR: {}".format(*param_evaluator.get_results()[1:]))
# logger.info("Diffusion - AP: {}, FPR: {}".format(*diff_evaluator.get_results()[1:]))
_, combined_ap, combined_fpr = evaluator.get_results()
logger.info(f"Results:")
logger.info(f"AP: {combined_ap:.4f}, FPR: {combined_fpr:.4f}")
def main(args):
# segmentation model
model_config = Config.fromfile(args.model_config)
model = init_model(model_config, device='cuda:0').eval()
model.load_state_dict(torch.load(args.segmentation_ckpt)['state_dict'])
feature_probe = get_feature_probe(model)
# diffusion model
diff_setup_hook_cfg = [h for h in model_config.custom_hooks if h.type=="DiffusionSetupHook"][0]
diffusion_model = get_diffusion_model(**diff_setup_hook_cfg).eval().cuda()
load_diffusion_checkpoint(diffusion_model, args.diffusion_ckpt)
diffusion_steps = get_denoising_steps_from_string(f"last{args.num_diffusion_steps}", args.diffusion_score_fn, total_steps=diffusion_model.diffusion_process.num_timesteps)
# data
ref_data_cfg = Config.fromfile(ref_data_cfg) if args.reference_data_config is not None else model_config
reference_loader = Runner.build_dataloader(ref_data_cfg.reference_dataloader)
test_loader = Runner.build_dataloader(Config.fromfile(args.test_data_config).test_dataloader)
# reference stats
logger.info(f"Getting reference statistics from {args.num_reference_samples} samples...")
reference_stats = get_reference_stats(model, diffusion_model, reference_loader.dataset, feature_probe,
args.num_reference_samples,
args.parametric_score_fn,
args.diffusion_score_fn,
diffusion_steps)
logger.info("Reference stats:")
logger.info(f"Parametric: mean = {reference_stats['param_mean_std_scores'][0]:.03f}, std = {reference_stats['param_mean_std_scores'][1]:.03f}")
logger.info(f"Diffusion: mean = {reference_stats['diff_mean_std_scores'][0]:.03f}, std = {reference_stats['diff_mean_std_scores'][1]:.03f}")
# evaluation
logger.info(f"OoD evaluation with {args.num_diffusion_steps} diffusion steps...")
eval_ood(model, diffusion_model, test_loader.dataset, feature_probe,
args.parametric_score_fn,
args.diffusion_score_fn,
diffusion_steps,
reference_stats["param_mean_std_scores"],
reference_stats["diff_mean_std_scores"],
save_dir=None, save_scoremaps=False,
dtype=torch.float16)
if __name__ == '__main__':
args = parse_args()
with warnings.catch_warnings():
warnings.filterwarnings("ignore", category=UserWarning)
warnings.filterwarnings("ignore", category=FutureWarning)
main(args)