Spaces:
Runtime error
Runtime error
Mehdi Cherti
commited on
Commit
•
06c5f0c
1
Parent(s):
8d2bdec
update available models
Browse files
run.py
CHANGED
@@ -1,8 +1,7 @@
|
|
1 |
import os
|
2 |
-
from clize import run
|
3 |
from glob import glob
|
4 |
from subprocess import call
|
5 |
-
|
6 |
def base():
|
7 |
return {
|
8 |
"slurm":{
|
@@ -34,7 +33,7 @@ def base():
|
|
34 |
"save_ckpt_every": 1,
|
35 |
"masked_mean": "",
|
36 |
"resume": "",
|
37 |
-
}
|
38 |
}
|
39 |
def ddgan_cc12m_v2():
|
40 |
cfg = base()
|
@@ -69,7 +68,6 @@ def ddgan_cc12m_v9():
|
|
69 |
cfg['model']['batch_size'] = 1
|
70 |
return cfg
|
71 |
|
72 |
-
|
73 |
def ddgan_cc12m_v11():
|
74 |
cfg = base()
|
75 |
cfg['model']['text_encoder'] = "google/t5-v1_1-large"
|
@@ -77,22 +75,78 @@ def ddgan_cc12m_v11():
|
|
77 |
cfg['model']['cross_attention'] = ""
|
78 |
return cfg
|
79 |
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
87 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
]
|
|
|
89 |
def get_model(model_name):
|
90 |
for model in models:
|
91 |
if model.__name__ == model_name:
|
92 |
return model()
|
93 |
|
94 |
|
95 |
-
def test(model_name, *, cond_text="", batch_size:int=None, epoch:int=None, guidance_scale:float=0, fid=False, real_img_dir=""):
|
96 |
|
97 |
cfg = get_model(model_name)
|
98 |
model = cfg['model']
|
@@ -104,6 +158,7 @@ def test(model_name, *, cond_text="", batch_size:int=None, epoch:int=None, guida
|
|
104 |
args = {}
|
105 |
args['exp'] = model_name
|
106 |
args['image_size'] = model['image_size']
|
|
|
107 |
args['num_channels'] = model['num_channels']
|
108 |
args['dataset'] = model['dataset']
|
109 |
args['num_channels_dae'] = model['num_channels_dae']
|
@@ -116,12 +171,35 @@ def test(model_name, *, cond_text="", batch_size:int=None, epoch:int=None, guida
|
|
116 |
args['text_encoder'] = model.get("text_encoder")
|
117 |
args['cross_attention'] = model.get("cross_attention")
|
118 |
args['guidance_scale'] = guidance_scale
|
|
|
|
|
|
|
119 |
|
120 |
if fid:
|
121 |
args['compute_fid'] = ''
|
122 |
args['real_img_dir'] = real_img_dir
|
123 |
-
|
|
|
124 |
print(cmd)
|
125 |
call(cmd, shell=True)
|
126 |
|
127 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
|
|
2 |
from glob import glob
|
3 |
from subprocess import call
|
4 |
+
import json
|
5 |
def base():
|
6 |
return {
|
7 |
"slurm":{
|
|
|
33 |
"save_ckpt_every": 1,
|
34 |
"masked_mean": "",
|
35 |
"resume": "",
|
36 |
+
},
|
37 |
}
|
38 |
def ddgan_cc12m_v2():
|
39 |
cfg = base()
|
|
|
68 |
cfg['model']['batch_size'] = 1
|
69 |
return cfg
|
70 |
|
|
|
71 |
def ddgan_cc12m_v11():
|
72 |
cfg = base()
|
73 |
cfg['model']['text_encoder'] = "google/t5-v1_1-large"
|
|
|
75 |
cfg['model']['cross_attention'] = ""
|
76 |
return cfg
|
77 |
|
78 |
+
def ddgan_cc12m_v12():
|
79 |
+
cfg = ddgan_cc12m_v11()
|
80 |
+
cfg['model']['text_encoder'] = "google/t5-v1_1-xl"
|
81 |
+
cfg['model']['preprocessing'] = 'random_resized_crop_v1'
|
82 |
+
return cfg
|
83 |
+
|
84 |
+
def ddgan_cc12m_v13():
|
85 |
+
cfg = ddgan_cc12m_v12()
|
86 |
+
cfg['model']['discr_type'] = "large_cond_attn"
|
87 |
+
return cfg
|
88 |
+
|
89 |
+
def ddgan_cc12m_v14():
|
90 |
+
cfg = ddgan_cc12m_v12()
|
91 |
+
cfg['model']['num_channels_dae'] = 192
|
92 |
+
return cfg
|
93 |
+
|
94 |
+
|
95 |
+
def ddgan_cifar10_cond17():
|
96 |
+
cfg = base()
|
97 |
+
cfg['model']['image_size'] = 32
|
98 |
+
cfg['model']['classifier_free_guidance_proba'] = 0.2
|
99 |
+
cfg['model']['ch_mult'] = "1 2 2 2"
|
100 |
+
cfg['model']['cross_attention'] = ""
|
101 |
+
cfg['model']['dataset'] = "cifar10"
|
102 |
+
cfg['model']['n_mlp'] = 4
|
103 |
+
return cfg
|
104 |
|
105 |
+
def ddgan_cifar10_cond18():
|
106 |
+
cfg = ddgan_cifar10_cond17()
|
107 |
+
cfg['model']['text_encoder'] = "google/t5-v1_1-xl"
|
108 |
+
return cfg
|
109 |
+
|
110 |
+
def ddgan_laion_aesthetic_v1():
|
111 |
+
cfg = ddgan_cc12m_v11()
|
112 |
+
cfg['model']['dataset_root'] = '"/p/scratch/ccstdl/cherti1/LAION-aesthetic/output/{00000..05038}.tar"'
|
113 |
+
return cfg
|
114 |
+
|
115 |
+
def ddgan_laion_aesthetic_v2():
|
116 |
+
cfg = ddgan_laion_aesthetic_v1()
|
117 |
+
cfg['model']['discr_type'] = "large_cond_attn"
|
118 |
+
return cfg
|
119 |
+
|
120 |
+
def ddgan_laion_aesthetic_v3():
|
121 |
+
cfg = ddgan_laion_aesthetic_v1()
|
122 |
+
cfg['model']['text_encoder'] = "google/t5-v1_1-xl"
|
123 |
+
return cfg
|
124 |
+
|
125 |
+
|
126 |
+
models = [
|
127 |
+
ddgan_cifar10_cond17, # cifar10, cross attn for discr
|
128 |
+
ddgan_cifar10_cond18, # cifar10, xl encoder
|
129 |
+
ddgan_cc12m_v2, # baseline (no large text encoder, no classifier guidance)
|
130 |
+
ddgan_cc12m_v6, # like v2 but using large T5 text encoder
|
131 |
+
ddgan_cc12m_v7, # like v2 but with classifier guidance
|
132 |
+
ddgan_cc12m_v8, # like v6 but classifier guidance
|
133 |
+
ddgan_cc12m_v9, # ~1B model but 64x64 resolution
|
134 |
+
ddgan_cc12m_v11, # large text encoder + cross attention + classifier free guidance
|
135 |
+
ddgan_cc12m_v12, # T5-XL + cross attention + classifier free guidance + random_resized_crop_v1
|
136 |
+
ddgan_cc12m_v13, # T5-XL + cross attention + classifier free guidance + random_resized_crop_v1 + cond attn
|
137 |
+
ddgan_cc12m_v14, # T5-XL + cross attention + classifier free guidance + random_resized_crop_v1 + 300M model
|
138 |
+
ddgan_laion_aesthetic_v1, # like ddgan_cc12m_v11 but fine-tuned on laion aesthetic
|
139 |
+
ddgan_laion_aesthetic_v2, # like ddgan_laion_aesthetic_v1 but trained from scratch with the new cross attn discr
|
140 |
+
ddgan_laion_aesthetic_v3, # like ddgan_laion_aesthetic_v1 but trained from scratch with T5-XL
|
141 |
]
|
142 |
+
|
143 |
def get_model(model_name):
|
144 |
for model in models:
|
145 |
if model.__name__ == model_name:
|
146 |
return model()
|
147 |
|
148 |
|
149 |
+
def test(model_name, *, cond_text="", batch_size:int=None, epoch:int=None, guidance_scale:float=0, fid=False, real_img_dir="", q=0.0, seed=0, nb_images_for_fid=0):
|
150 |
|
151 |
cfg = get_model(model_name)
|
152 |
model = cfg['model']
|
|
|
158 |
args = {}
|
159 |
args['exp'] = model_name
|
160 |
args['image_size'] = model['image_size']
|
161 |
+
args['seed'] = seed
|
162 |
args['num_channels'] = model['num_channels']
|
163 |
args['dataset'] = model['dataset']
|
164 |
args['num_channels_dae'] = model['num_channels_dae']
|
|
|
171 |
args['text_encoder'] = model.get("text_encoder")
|
172 |
args['cross_attention'] = model.get("cross_attention")
|
173 |
args['guidance_scale'] = guidance_scale
|
174 |
+
args['masked_mean'] = model.get("masked_mean")
|
175 |
+
args['dynamic_thresholding_quantile'] = q
|
176 |
+
args['n_mlp'] = model.get("n_mlp")
|
177 |
|
178 |
if fid:
|
179 |
args['compute_fid'] = ''
|
180 |
args['real_img_dir'] = real_img_dir
|
181 |
+
args['nb_images_for_fid'] = nb_images_for_fid
|
182 |
+
cmd = "python -u test_ddgan.py " + " ".join(f"--{k} {v}" for k, v in args.items() if v is not None)
|
183 |
print(cmd)
|
184 |
call(cmd, shell=True)
|
185 |
|
186 |
+
def eval_results(model_name):
|
187 |
+
import pandas as pd
|
188 |
+
rows = []
|
189 |
+
cfg = get_model(model_name)
|
190 |
+
model = cfg['model']
|
191 |
+
paths = glob('./saved_info/dd_gan/{}/{}/fid*.json'.format(model["dataset"], model_name))
|
192 |
+
for path in paths:
|
193 |
+
with open(path, "r") as fd:
|
194 |
+
data = json.load(fd)
|
195 |
+
row = {}
|
196 |
+
row['fid'] = data['fid']
|
197 |
+
row['epoch'] = data['epoch_id']
|
198 |
+
rows.append(row)
|
199 |
+
out = './saved_info/dd_gan/{}/{}/fid.csv'.format(model["dataset"], model_name)
|
200 |
+
df = pd.DataFrame(rows)
|
201 |
+
df.to_csv(out, index=False)
|
202 |
+
|
203 |
+
if __name__ == "__main__":
|
204 |
+
from clize import run
|
205 |
+
run([test, eval_results])
|