-
Notifications
You must be signed in to change notification settings - Fork 0
/
config.py
257 lines (229 loc) · 16.7 KB
/
config.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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
from dataclasses import dataclass, field
from typing import List
@dataclass
class SamplingStatic:
NUM_INFERENCE_STEPS: int = 25
SHOW_PROMPT_N: int = 5
MAX_BATCH_N: int = 9
GUIDANCE_SCALE: float = 7.5
IMAGE_NUM_PER_PROMPT: int = 1
IMAGE_NUM_PER_GRID_SAMPLE: int = 9
FORMAT: str = "png"
CLEAN_BACKDOOR_BOTH: str = 'bc'
CLEAN_BACKDOOR_CLEAN: str = 'c'
CLEAN_BACKDOOR_BACKDOOR: str = 'b'
TRIG_START_POS: int = -1
TRIG_END_POS: int = -1
SEED: int = 1
HANDLE_FN: callable = lambda *arg: None
HANDLE_BATCH_FN: callable = lambda *arg: None
FORCE_REGENERATE: bool = False
# @dataclass
# class SamplingConfig:
# base_path: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': True, 'help': "Path to trained model"}))
# ckpt_step: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': None, 'required': False, 'help': "Checkpoint training step"}))
# prompt: str=None
# clean_backdoor: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': SamplingStatic.CLEAN_BACKDOOR_BOTH, 'required': False, 'help': "Sample clean or backdoor images"}))
# dataset_name: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': False, 'help': "Training dataset for backdooring"}))
# image_trigger: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': False, 'help': "Image trigger"}))
# caption_trigger: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': False, 'help': "Caption trigger"}))
# max_batch_n: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': 9, 'required': False, 'help': "Sampling batch size"}))
# img_num: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': 9, 'required': False, 'help': "Image grid size"}))
# num_inference_steps: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': 50, 'required': False, 'help': "Number of sampling steps"}))
# guidance_scale: float = field(default_factory=lambda: ({'export': True, 'type': int, 'default': 7.5, 'required': False, 'help': "Scale of conditional guidance"}))
# enable_lora: bool = field(default_factory=lambda: ({'export': True, 'default': False, 'action': "store_true", 'help': "Enable LoRA"}))
# gpu: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': '0', 'required': False, 'help': "Used GPU ID"}))
# sampling_config_file: str = 'sampling.json'
# train_config_file: str = 'train.json'
# format: str = "png"
# seed: int=1
# args_key: str = 'args'
# default_key: str = 'default'
# final_key: str = 'final'
@dataclass
class SamplingConfig:
base_path: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': True, 'help': "Path to trained model"}))
ckpt_step: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': None, 'required': False, 'help': "Checkpoint training step"}))
# prompt: str = None
# clean_backdoor: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': SamplingStatic.CLEAN_BACKDOOR_BOTH, 'required': False, 'help': "Sample clean or backdoor images"}))
dataset_name: str = None
# in_dist_ds: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "In-distribution dataset name"}))
in_dist_ds: str = None
# out_dist_ds: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Out-distribution dataset name"}))
out_dist_ds: str = None
# target: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Backdoor target"}))
target: str = None
# image_trigger: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Image trigger"}))
image_trigger: str = None
# caption_trigger: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Caption trigger"}))
caption_trigger: str = None
max_batch_n: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': SamplingStatic.MAX_BATCH_N, 'required': False, 'help': "Sampling batch size"}))
# max_batch_n: int = 9
sched: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': ModelSchedStatic.SCHED, 'required': False, 'help': "Sampler type"}))
num_inference_steps: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': SamplingStatic.NUM_INFERENCE_STEPS, 'required': False, 'help': "Number of sampling steps"}))
# num_inference_steps: int = 50
guidance_scale: float = field(default_factory=lambda: ({'export': False, 'type': float, 'default': SamplingStatic.GUIDANCE_SCALE, 'required': False, 'help': "Scale of conditional guidance"}))
# guidance_scale: float = 7.5
use_lora: bool = True
# enable_lora: bool = field(default_factory=lambda: ({'export': True, 'default': False, 'action': "store_true", 'help': "Enable LoRA"}))
gpu: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': '0', 'required': False, 'help': "Used GPU ID"}))
trig_start_pos: int = SamplingStatic.TRIG_START_POS
trig_end_pos: int = SamplingStatic.TRIG_END_POS
img_num_per_grid_sample: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': SamplingStatic.IMAGE_NUM_PER_GRID_SAMPLE, 'required': False, 'help': "Number of samples for every prompt"}))
force_regenerate: bool = field(default_factory=lambda: ({'export': True, 'action': action_generator(SamplingStatic.FORCE_REGENERATE), 'help': "Regenerate samples or not"}))
lora_base_model: str = "CompVis/stable-diffusion-v1-4"
ds_base_path: str = 'datasets'
sampling_config_file: str = 'sampling.json'
train_config_file: str = 'args.json'
format: str = "png"
seed: int = SamplingStatic.SEED
args_key: str = 'args'
default_key: str = 'default'
final_key: str = 'final'
DEFAULT_PROMPTS_POKEMON: List[str] = [
"a photo of cat",
"a photo of dog",
"Grunge Dallas skyline with American flag illustration",
"a drawing of a pikachu with a green leaf on its head",
"a blue and white bird with its wings spread",
"a cartoon character with a cat like body",
"a drawing of a green pokemon with red eyes",
"a drawing of a pikachu with a green leaf on its head",
"A collage of images with various slogans.",
"The American flag and a city skyline.",
"An advertisement for the new Owlly Night Owls.",
]
DEFAULT_PROMPTS_CELEBA: List[str] = [
"a photo of cat",
"a photo of dog",
"This woman is in the thirties and has no glasses, and a big smile with her mouth a bit open. This lady has no bangs at all.', 'Bangs': 'Her whole forehead is visible.",
"This young girl has no fringe, a smile, and no glasses.",
"This gentleman has stubble. This man looks very young and has no glasses, no smile, and no bangs.",
"This guy doesn't have any beard at all. This man is in his thirties and has no smile, and no glasses. The whole forehead is visible without any fringe.",
"This man has thin frame sunglasses. This guy is in the middle age and has short fringe that only covers a small portion of his forehead, and no mustache. He has a beaming face.",
"This person has no fringe, and a extremely mild smile. This lady is a teen and has no eyeglasses.",
"This female has no eyeglasses, and no bangs. This person is in the thirties and has a mild smile.",
"A collage of images with various slogans.",
"The American flag and a city skyline.",
"An advertisement for the new Owlly Night Owls.",
]
@dataclass
class PromptDatasetStatic:
FORCE_UPDATE: bool = False
IN_DIST: str = "IN_DIST"
OUT_DIST: str = "OUT_DIST"
DEFAULT_DIST: str = "NONE_DIST"
TRAIN_SPLIT: str = "TRAIN_SPLIT"
TEST_SPLIT: str = "TEST_SPLIT"
FULL_SPLIT: str = "FULL_SPLIT"
DEFAULT_SPLIT: str = "NONE_SPLIT"
IN_DIST_NAME: str = "IN"
OUT_DIST_NAME: str = "OUT"
OUT_DIST_SAMPLE_N: int = 800
TRAIN_SPLIT_NAME: str = "TRAIN"
TEST_SPLIT_NAME: str = "TEST"
FULL_SPLIT_NAME: str = "FULL"
TRAIN_SPLIT_RATIO: int = 90
@dataclass
class ModelSchedStatic:
# PNDM_SCHED: str = "PNDM_SCHED"
DPM_SOLVER_PP_O2_SCHED: str = "DPM_SOLVER_PP_O2_SCHED"
SCHED: str = DPM_SOLVER_PP_O2_SCHED
@dataclass
class MeasuringStatic:
IN_DIST_TRAIN_DIR: str = 'in_dist_train'
IN_DIST_TEST_DIR: str = 'in_dist_test'
IN_DIST_FULL_DIR: str = 'in_dist_full'
OUT_DIST_FULL_DIR: str = 'out_dist_full'
OUT_DIST_DIR: str = 'out_dist'
IN_DIST_TRAIN_CLEAN_SAMPLE_DIR: str = f'{IN_DIST_TRAIN_DIR}_clean_sample'
IN_DIST_TRAIN_CAPTION_BACKDOOR_SAMPLE_DIR: str = f'{IN_DIST_TRAIN_DIR}_caption_backdoor_sample'
IN_DIST_TRAIN_IMAGE_BACKDOOR_SAMPLE_DIR: str = f'{IN_DIST_TRAIN_DIR}_image_backdoor_sample'
IN_DIST_TEST_CLEAN_SAMPLE_DIR: str = f'{IN_DIST_TEST_DIR}_clean_sample'
IN_DIST_TEST_CAPTION_BACKDOOR_SAMPLE_DIR: str = f'{IN_DIST_TEST_DIR}_caption_backdoor_sample'
IN_DIST_TEST_IMAGE_BACKDOOR_SAMPLE_DIR: str = f'{IN_DIST_TEST_DIR}_image_backdoor_sample'
OUT_DIST_CLEAN_SAMPLE_DIR: str = f'{OUT_DIST_DIR}_clean_sample'
OUT_DIST_CAPTION_BACKDOOR_SAMPLE_DIR: str = f'{OUT_DIST_DIR}_caption_backdoor_sample'
OUT_DIST_IMAGE_BACKDOOR_SAMPLE_DIR: str = f'{OUT_DIST_DIR}_image_backdoor_sample'
IMAGE_BACKDOOR: str = 'image_backdoor'
CAPTION_BACKDOOR: str = 'caption_backdoor'
CLEAN: str = 'clean'
FORMAT: str = SamplingStatic.FORMAT
DIR_NAME: str = "measuring_cache"
# Measuring Options
MEASURING_CLEAN: str = "measuring_clean"
MEASURING_BACKDOOR: str = "measuring_backdoor"
METRIC_FID: str = "METRIC_FID"
METRIC_MSE: str = "METRIC_MSE"
METRIC_SSIM: str = "METRIC_SSIM"
METRIC_MSE_THRES: float = 0.1
MAX_BATCH_N: int = 9
FID_MAX_BATCH_N: int = 64
IMAGE_NUM_PER_PROMPT: int = 1
IMAGE_NUM_PER_GRID_SAMPLE: int = 9
DEFAULT_SAMPLE_PROMPTS_N: int = 20
# MAX_MEASURING_SAMPLES: int = 33
MAX_MEASURING_SAMPLES: int = 1000
# MAX_MEASURING_SAMPLES: int = 3000
# MAX_MEASURING_SAMPLES: int = 5
FORCE_REGENERATE: bool = SamplingStatic.FORCE_REGENERATE
DEVICE: str = "cuda:0"
SCORE_FILE: str = "score.json"
SEED: int = SamplingStatic.SEED
@dataclass
class MeasuringConfig:
base_path: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': True, 'help': "Path to trained model"}))
# base_path: str = "lora8/res_POKEMON-CAPTION_NONE-TRIGGER_EMOJI_SOCCER-HACKER_pr1.0_ca0_caw1.0_rctp0_lr0.0001_step50000_prior1.0_lora4_new-set"
ckpt_step: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': -1, 'required': False, 'help': "Checkpoint training step"}))
# ckpt_step: int = -1
project: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': 'Default', 'required': True, 'help': "Wandb project name"}))
# project: str = 'Default'
# clean_backdoor: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': SamplingStatic.CLEAN_BACKDOOR_BOTH, 'required': False, 'help': "Sample clean or backdoor images"}))
dataset_name: str = None
# in_dist_ds: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "In-distribution dataset name"}))
in_dist_ds: str = None
# out_dist_ds: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Out-distribution dataset name"}))
out_dist_ds: str = None
# target: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Backdoor target"}))
target: str = None
# image_trigger: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Image trigger"}))
image_trigger: str = None
# caption_trigger: str = field(default_factory=lambda: ({'export': False, 'type': str, 'default': None, 'required': False, 'help': "Caption trigger"}))
caption_trigger: str = None
max_batch_n: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': MeasuringStatic.MAX_BATCH_N, 'required': False, 'help': "Sampling batch size"}))
fid_max_batch_n: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': MeasuringStatic.FID_MAX_BATCH_N, 'required': False, 'help': "FID batch size"}))
# max_batch_n: int = 9
sched: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': ModelSchedStatic.SCHED, 'required': False, 'help': "Sampler type"}))
num_inference_steps: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': SamplingStatic.NUM_INFERENCE_STEPS, 'required': False, 'help': "Number of sampling steps"}))
# num_inference_steps: int = 50
guidance_scale: float = field(default_factory=lambda: ({'export': False, 'type': float, 'default': SamplingStatic.GUIDANCE_SCALE, 'required': False, 'help': "Scale of conditional guidance"}))
# guidance_scale: float = 7.5
use_lora: bool = True
gpu: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': '0', 'required': False, 'help': "Used GPU ID"}))
# gpu: str = '0'
# Measure on In/Out Distribution, Training/Testing Split, Clean/Backdoor
mode: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': f"{PromptDatasetStatic.DEFAULT_DIST}|{PromptDatasetStatic.DEFAULT_SPLIT}|{MeasuringStatic.MEASURING_CLEAN}", 'required': False, 'help': "Measure in/out distribution, train/test split, clean/backdoor dataset"}))
# in_out_dist: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': None, 'required': PromptDatasetStatic.DEFAULT_DIST, 'help': "Measure in/out distribution dataset"}))
# train_test_split: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': PromptDatasetStatic.DEFAULT_SPLIT, 'required': False, 'help': "Measure training/testing split"}))
# clean_backdoor: str = field(default_factory=lambda: ({'export': True, 'type': str, 'default': MeasuringStatic.MEASURING_CLEAN, 'required': False, 'help': "Measure clean/backdoor samples"}))
mse_thres: int = MeasuringStatic.METRIC_MSE_THRES
trig_start_pos: int = SamplingStatic.TRIG_START_POS
trig_end_pos: int = SamplingStatic.TRIG_END_POS
image_num_per_prompt: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': MeasuringStatic.IMAGE_NUM_PER_PROMPT, 'required': False, 'help': "Number of samples for every prompt"}))
img_num_per_grid_sample: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': MeasuringStatic.IMAGE_NUM_PER_GRID_SAMPLE, 'required': False, 'help': "Number of images for every image grid"}))
max_measuring_samples: int = field(default_factory=lambda: ({'export': True, 'type': int, 'default': MeasuringStatic.MAX_MEASURING_SAMPLES, 'required': False, 'help': "Number of generative images for the evaluation"}))
force_regenerate: bool = field(default_factory=lambda: ({'export': True, 'action': action_generator(MeasuringStatic.FORCE_REGENERATE), 'help': "Regenerate samples or not"}))
lora_base_model: str = "CompVis/stable-diffusion-v1-4"
ds_base_path: str = 'datasets'
measure_config_file: str = 'measure.json'
train_config_file: str = 'args.json'
format: str = "png"
seed: int = MeasuringStatic.SEED
args_key: str = 'args'
default_key: str = 'default'
final_key: str = 'final'
caption_similarity: float = None
def action_generator(default_action: bool):
if default_action:
return 'store_false'
return 'store_true'