Skip to content

Commit 81875d5

Browse files
rwightmanrwightman
committed
Update sotabench model list, add Mean-Max pooling DPN variants, disable download progress
1 parent b06dce8 commit 81875d5

File tree

2 files changed

+125
-191
lines changed

2 files changed

+125
-191
lines changed

‎sotabench.py‎

Lines changed: 124 additions & 190 deletions
Original file line numberDiff line numberDiff line change
@@ -1,219 +1,153 @@
11
from torchbench.image_classification import ImageNet
2-
from timm import create_model, list_models
2+
from timm import create_model
33
from timm.data import resolve_data_config, create_transform
4+
from timm.models import TestTimePoolHead
45
import os
56

67
NUM_GPU = 1
78
BATCH_SIZE = 256 * NUM_GPU
89

910

10-
def _attrib(paper_model_name='', paper_arxiv_id='', batch_size=BATCH_SIZE):
11+
def _entry(model_name, paper_model_name, paper_arxiv_id, batch_size=BATCH_SIZE, ttp=False, args=dict()):
1112
return dict(
13+
model=model_name,
1214
paper_model_name=paper_model_name,
1315
paper_arxiv_id=paper_arxiv_id,
14-
batch_size=batch_size)
16+
batch_size=batch_size,
17+
ttp=ttp,
18+
args=args)
1519

16-
model_map = dict(
17-
#adv_inception_v3=_attrib(paper_model_name='Adversarial Inception V3', paper_arxiv_id=),
18-
#densenet121=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
19-
#densenet161=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
20-
#densenet169=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
21-
#densenet201=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
22-
dpn68=_attrib(
23-
paper_model_name='DPN-68', paper_arxiv_id='1707.01629'),
24-
dpn68b=_attrib(
25-
paper_model_name='DPN-68b', paper_arxiv_id='1707.01629'),
26-
dpn92=_attrib(
27-
paper_model_name='DPN-92', paper_arxiv_id='1707.01629'),
28-
dpn98=_attrib(
29-
paper_model_name='DPN-98', paper_arxiv_id='1707.01629'),
30-
dpn107=_attrib(
31-
paper_model_name='DPN-107', paper_arxiv_id='1707.01629'),
32-
dpn131=_attrib(
33-
paper_model_name='DPN-131', paper_arxiv_id='1707.01629'),
34-
efficientnet_b0=_attrib(
35-
paper_model_name='EfficientNet-B0', paper_arxiv_id='1905.11946'),
36-
efficientnet_b1=_attrib(
37-
paper_model_name='EfficientNet-B1', paper_arxiv_id='1905.11946'),
38-
efficientnet_b2=_attrib(
39-
paper_model_name='EfficientNet-B2', paper_arxiv_id='1905.11946'),
40-
#ens_adv_inception_resnet_v2=_attrib(paper_model_name=, paper_arxiv_id=),
41-
fbnetc_100=_attrib(
42-
paper_model_name='FBNet-C', paper_arxiv_id='1812.03443'),
43-
gluon_inception_v3=_attrib(
44-
paper_model_name='Inception V3', paper_arxiv_id='1512.00567'),
45-
gluon_resnet18_v1b=_attrib(
46-
paper_model_name='ResNet-18', paper_arxiv_id='1812.01187'),
47-
gluon_resnet34_v1b=_attrib(
48-
paper_model_name='ResNet-34', paper_arxiv_id='1812.01187'),
49-
gluon_resnet50_v1b=_attrib(
50-
paper_model_name='ResNet-50', paper_arxiv_id='1812.01187'),
51-
gluon_resnet50_v1c=_attrib(
52-
paper_model_name='ResNet-50-C', paper_arxiv_id='1812.01187'),
53-
gluon_resnet50_v1d=_attrib(
54-
paper_model_name='ResNet-50-D', paper_arxiv_id='1812.01187'),
55-
gluon_resnet50_v1s=_attrib(
56-
paper_model_name='ResNet-50-S', paper_arxiv_id='1812.01187'),
57-
gluon_resnet101_v1b=_attrib(
58-
paper_model_name='ResNet-101', paper_arxiv_id='1812.01187'),
59-
gluon_resnet101_v1c=_attrib(
60-
paper_model_name='ResNet-101-C', paper_arxiv_id='1812.01187'),
61-
gluon_resnet101_v1d=_attrib(
62-
paper_model_name='ResNet-101-D', paper_arxiv_id='1812.01187'),
63-
gluon_resnet101_v1s=_attrib(
64-
paper_model_name='ResNet-101-S', paper_arxiv_id='1812.01187'),
65-
gluon_resnet152_v1b=_attrib(
66-
paper_model_name='ResNet-152', paper_arxiv_id='1812.01187'),
67-
gluon_resnet152_v1c=_attrib(
68-
paper_model_name='ResNet-152-C', paper_arxiv_id='1812.01187'),
69-
gluon_resnet152_v1d=_attrib(
70-
paper_model_name='ResNet-152-D', paper_arxiv_id='1812.01187'),
71-
gluon_resnet152_v1s=_attrib(
72-
paper_model_name='ResNet-152-S', paper_arxiv_id='1812.01187'),
73-
gluon_resnext50_32x4d=_attrib(
74-
paper_model_name='ResNeXt-50 32x4d', paper_arxiv_id='1812.01187'),
75-
gluon_resnext101_32x4d=_attrib(
76-
paper_model_name='ResNeXt-101 32x4d', paper_arxiv_id='1812.01187'),
77-
gluon_resnext101_64x4d=_attrib(
78-
paper_model_name='ResNeXt-101 64x4d', paper_arxiv_id='1812.01187'),
79-
gluon_senet154=_attrib(
80-
paper_model_name='SENet-154', paper_arxiv_id='1812.01187'),
81-
gluon_seresnext50_32x4d=_attrib(
82-
paper_model_name='SE-ResNeXt-50 32x4d', paper_arxiv_id='1812.01187'),
83-
gluon_seresnext101_32x4d=_attrib(
84-
paper_model_name='SE-ResNeXt-101 32x4d', paper_arxiv_id='1812.01187'),
85-
gluon_seresnext101_64x4d=_attrib(
86-
paper_model_name='SE-ResNeXt-101 64x4d', paper_arxiv_id='1812.01187'),
87-
gluon_xception65=_attrib(
88-
paper_model_name='Modified Aligned Xception', paper_arxiv_id='1802.02611', batch_size=BATCH_SIZE//2),
89-
ig_resnext101_32x8d=_attrib(
90-
paper_model_name='ResNeXt-101 32x8d', paper_arxiv_id='1805.00932'),
91-
ig_resnext101_32x16d=_attrib(
92-
paper_model_name='ResNeXt-101 32x16d', paper_arxiv_id='1805.00932'),
93-
ig_resnext101_32x32d=_attrib(
94-
paper_model_name='ResNeXt-101 32x32d', paper_arxiv_id='1805.00932', batch_size=BATCH_SIZE//2),
95-
ig_resnext101_32x48d=_attrib(
96-
paper_model_name='ResNeXt-101 32x48d', paper_arxiv_id='1805.00932', batch_size=BATCH_SIZE//4),
97-
inception_resnet_v2=_attrib(
98-
paper_model_name='Inception ResNet V2', paper_arxiv_id='1602.07261'),
99-
#inception_v3=dict(paper_model_name='Inception V3', paper_arxiv_id=), # same weights as torchvision
100-
inception_v4=_attrib(
101-
paper_model_name='Inception V4', paper_arxiv_id='1602.07261'),
102-
mixnet_l=_attrib(
103-
paper_model_name='MixNet-L', paper_arxiv_id='1907.09595'),
104-
mixnet_m=_attrib(
105-
paper_model_name='MixNet-M', paper_arxiv_id='1907.09595'),
106-
mixnet_s=_attrib(
107-
paper_model_name='MixNet-S', paper_arxiv_id='1907.09595'),
108-
mnasnet_100=_attrib(
109-
paper_model_name='MnasNet-B1', paper_arxiv_id='1807.11626'),
110-
mobilenetv3_100=_attrib(
111-
paper_model_name='MobileNet V3(1.0)', paper_arxiv_id='1905.02244'),
112-
nasnetalarge=_attrib(
113-
paper_model_name='NASNet-A Large', paper_arxiv_id='1707.07012', batch_size=BATCH_SIZE//4),
114-
pnasnet5large=_attrib(
115-
paper_model_name='PNASNet-5', paper_arxiv_id='1712.00559', batch_size=BATCH_SIZE//4),
116-
resnet18=_attrib(
117-
paper_model_name='ResNet-18', paper_arxiv_id='1812.01187'),
118-
resnet26=_attrib(
119-
paper_model_name='ResNet-26', paper_arxiv_id='1812.01187'),
120-
resnet26d=_attrib(
121-
paper_model_name='ResNet-26-D', paper_arxiv_id='1812.01187'),
122-
resnet34=_attrib(
123-
paper_model_name='ResNet-34', paper_arxiv_id='1812.01187'),
124-
resnet50=_attrib(
125-
paper_model_name='ResNet-50', paper_arxiv_id='1812.01187'),
126-
#resnet101=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
127-
#resnet152=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
128-
resnext50_32x4d=_attrib(
129-
paper_model_name='ResNeXt-50 32x4d', paper_arxiv_id='1812.01187'),
130-
resnext50d_32x4d=_attrib(
131-
paper_model_name='ResNeXt-50-D 32x4d', paper_arxiv_id='1812.01187'),
132-
#resnext101_32x8d=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
133-
semnasnet_100=_attrib(
134-
paper_model_name='MnasNet-A1', paper_arxiv_id='1807.11626'),
135-
senet154=_attrib(
136-
paper_model_name='SENet-154', paper_arxiv_id='1709.01507'),
137-
seresnet18=_attrib(
138-
paper_model_name='SE-ResNet-18', paper_arxiv_id='1709.01507'),
139-
seresnet34=_attrib(
140-
paper_model_name='SE-ResNet-34', paper_arxiv_id='1709.01507'),
141-
seresnet50=_attrib(
142-
paper_model_name='SE-ResNet-50', paper_arxiv_id='1709.01507'),
143-
seresnet101=_attrib(
144-
paper_model_name='SE-ResNet-101', paper_arxiv_id='1709.01507'),
145-
seresnet152=_attrib(
146-
paper_model_name='SE-ResNet-152', paper_arxiv_id='1709.01507'),
147-
seresnext26_32x4d=_attrib(
148-
paper_model_name='SE-ResNeXt-26 32x4d', paper_arxiv_id='1709.01507'),
149-
seresnext50_32x4d=_attrib(
150-
paper_model_name='SE-ResNeXt-50 32x4d', paper_arxiv_id='1709.01507'),
151-
seresnext101_32x4d=_attrib(
152-
paper_model_name='SE-ResNeXt-101 32x4d', paper_arxiv_id='1709.01507'),
153-
spnasnet_100=_attrib(
154-
paper_model_name='Single-Path NAS', paper_arxiv_id='1904.02877'),
155-
tf_efficientnet_b0=_attrib(
156-
paper_model_name='EfficientNet-B0', paper_arxiv_id='1905.11946'),
157-
tf_efficientnet_b1=_attrib(
158-
paper_model_name='EfficientNet-B1', paper_arxiv_id='1905.11946'),
159-
tf_efficientnet_b2=_attrib(
160-
paper_model_name='EfficientNet-B2', paper_arxiv_id='1905.11946'),
161-
tf_efficientnet_b3=_attrib(
162-
paper_model_name='EfficientNet-B3', paper_arxiv_id='1905.11946', batch_size=BATCH_SIZE//2),
163-
tf_efficientnet_b4=_attrib(
164-
paper_model_name='EfficientNet-B4', paper_arxiv_id='1905.11946', batch_size=BATCH_SIZE//2),
165-
tf_efficientnet_b5=_attrib(
166-
paper_model_name='EfficientNet-B5', paper_arxiv_id='1905.11946', batch_size=BATCH_SIZE//4),
167-
tf_efficientnet_b6=_attrib(
168-
paper_model_name='EfficientNet-B6', paper_arxiv_id='1905.11946', batch_size=BATCH_SIZE//8),
169-
tf_efficientnet_b7=_attrib(
170-
paper_model_name='EfficientNet-B7', paper_arxiv_id='1905.11946', batch_size=BATCH_SIZE//8),
171-
tf_efficientnet_es=_attrib(
172-
paper_model_name='EfficientNet-EdgeTPU-S', paper_arxiv_id='1905.11946'),
173-
tf_efficientnet_em=_attrib(
174-
paper_model_name='EfficientNet-EdgeTPU-M', paper_arxiv_id='1905.11946'),
175-
tf_efficientnet_el=_attrib(
176-
paper_model_name='EfficientNet-EdgeTPU-L', paper_arxiv_id='1905.11946', batch_size=BATCH_SIZE//2),
177-
tf_inception_v3=_attrib(
178-
paper_model_name='Inception V3', paper_arxiv_id='1512.00567'),
179-
tf_mixnet_l=_attrib(
180-
paper_model_name='MixNet-L', paper_arxiv_id='1907.09595'),
181-
tf_mixnet_m=_attrib(
182-
paper_model_name='MixNet-M', paper_arxiv_id='1907.09595'),
183-
tf_mixnet_s=_attrib(
184-
paper_model_name='MixNet-S', paper_arxiv_id='1907.09595'),
185-
#tv_resnet34=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
186-
#tv_resnet50=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
187-
#tv_resnext50_32x4d=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
188-
#wide_resnet50_2=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
189-
#wide_resnet101_2=_attrib(paper_model_name=, paper_arxiv_id=), # same weights as torchvision
190-
xception=_attrib(
191-
paper_model_name='Xception', paper_arxiv_id='1610.02357'),
192-
)
19320

194-
model_names = list_models(pretrained=True)
21+
model_list = [
22+
#_entry('adv_inception_v3', 'Adversarial Inception V3', ),
23+
#_entry('densenet121'), # same weights as torchvision
24+
#_entry('densenet161'), # same weights as torchvision
25+
#_entry('densenet169'), # same weights as torchvision
26+
#_entry('densenet201'), # same weights as torchvision
27+
_entry('dpn68', 'DPN-68 (224x224)', '1707.01629'),
28+
_entry('dpn68b', 'DPN-68b (224x224)', '1707.01629'),
29+
_entry('dpn92', 'DPN-92 (224x224)', '1707.01629'),
30+
_entry('dpn98', 'DPN-98 (224x224)', '1707.01629'),
31+
_entry('dpn107', 'DPN-107 (224x224)', '1707.01629'),
32+
_entry('dpn131', 'DPN-131 (224x224)', '1707.01629'),
33+
_entry('dpn68', 'DPN-68 (320x320, Mean-Max Pooling)', '1707.01629', ttp=True, args=dict(img_size=320)),
34+
_entry('dpn68b', 'DPN-68b (320x320, Mean-Max Pooling)', '1707.01629', ttp=True, args=dict(img_size=320)),
35+
_entry('dpn92', 'DPN-92 (320x320, Mean-Max Pooling)', '1707.01629',
36+
ttp=True, args=dict(img_size=320), batch_size=BATCH_SIZE//2),
37+
_entry('dpn98', 'DPN-98 (320x320, Mean-Max Pooling)', '1707.01629',
38+
ttp=True, args=dict(img_size=320), batch_size=BATCH_SIZE//2),
39+
_entry('dpn107', 'DPN-107 (320x320, Mean-Max Pooling)', '1707.01629',
40+
ttp=True, args=dict(img_size=320), batch_size=BATCH_SIZE//4),
41+
_entry('dpn131', 'DPN-131 (320x320, Mean-Max Pooling)', '1707.01629',
42+
ttp=True, args=dict(img_size=320), batch_size=BATCH_SIZE//4),
43+
_entry('efficientnet_b0', 'EfficientNet-B0', '1905.11946'),
44+
_entry('efficientnet_b1', 'EfficientNet-B1', '1905.11946'),
45+
_entry('efficientnet_b2', 'EfficientNet-B2', '1905.11946'),
46+
#_entry('ens_adv_inception_resnet_v2', 'Ensemble Adversarial Inception V3'),
47+
_entry('fbnetc_100', 'FBNet-C', '1812.03443'),
48+
_entry('gluon_inception_v3', 'Inception V3', '1512.00567'),
49+
_entry('gluon_resnet18_v1b', 'ResNet-18', '1812.01187'),
50+
_entry('gluon_resnet34_v1b', 'ResNet-34', '1812.01187'),
51+
_entry('gluon_resnet50_v1b', 'ResNet-50', '1812.01187'),
52+
_entry('gluon_resnet50_v1c', 'ResNet-50-C', '1812.01187'),
53+
_entry('gluon_resnet50_v1d', 'ResNet-50-D', '1812.01187'),
54+
_entry('gluon_resnet50_v1s', 'ResNet-50-S', '1812.01187'),
55+
_entry('gluon_resnet101_v1b', 'ResNet-101', '1812.01187'),
56+
_entry('gluon_resnet101_v1c', 'ResNet-101-C', '1812.01187'),
57+
_entry('gluon_resnet101_v1d', 'ResNet-101-D', '1812.01187'),
58+
_entry('gluon_resnet101_v1s', 'ResNet-101-S', '1812.01187'),
59+
_entry('gluon_resnet152_v1b', 'ResNet-152', '1812.01187'),
60+
_entry('gluon_resnet152_v1c', 'ResNet-152-C', '1812.01187'),
61+
_entry('gluon_resnet152_v1d', 'ResNet-152-D', '1812.01187'),
62+
_entry('gluon_resnet152_v1s', 'ResNet-152-S', '1812.01187'),
63+
_entry('gluon_resnext50_32x4d', 'ResNeXt-50 32x4d', '1812.01187'),
64+
_entry('gluon_resnext101_32x4d', 'ResNeXt-101 32x4d', '1812.01187'),
65+
_entry('gluon_resnext101_64x4d', 'ResNeXt-101 64x4d', '1812.01187'),
66+
_entry('gluon_senet154', 'SENet-154', '1812.01187'),
67+
_entry('gluon_seresnext50_32x4d', 'SE-ResNeXt-50 32x4d', '1812.01187'),
68+
_entry('gluon_seresnext101_32x4d', 'SE-ResNeXt-101 32x4d', '1812.01187'),
69+
_entry('gluon_seresnext101_64x4d', 'SE-ResNeXt-101 64x4d', '1812.01187'),
70+
_entry('gluon_xception65', 'Modified Aligned Xception', '1802.02611', batch_size=BATCH_SIZE//2),
71+
_entry('ig_resnext101_32x8d', 'ResNeXt-101 32x8d', '1805.00932'),
72+
_entry('ig_resnext101_32x16d', 'ResNeXt-101 32x16d', '1805.00932'),
73+
_entry('ig_resnext101_32x32d', 'ResNeXt-101 32x32d', '1805.00932', batch_size=BATCH_SIZE//2),
74+
_entry('ig_resnext101_32x48d', 'ResNeXt-101 32x48d', '1805.00932', batch_size=BATCH_SIZE//4),
75+
_entry('inception_resnet_v2', 'Inception ResNet V2', '1602.07261'),
76+
#_entry('inception_v3', paper_model_name='Inception V3', ), # same weights as torchvision
77+
_entry('inception_v4', 'Inception V4', '1602.07261'),
78+
_entry('mixnet_l', 'MixNet-L', '1907.09595'),
79+
_entry('mixnet_m', 'MixNet-M', '1907.09595'),
80+
_entry('mixnet_s', 'MixNet-S', '1907.09595'),
81+
_entry('mnasnet_100', 'MnasNet-B1', '1807.11626'),
82+
_entry('mobilenetv3_100', 'MobileNet V3(1.0)', '1905.02244'),
83+
_entry('nasnetalarge', 'NASNet-A Large', '1707.07012', batch_size=BATCH_SIZE//4),
84+
_entry('pnasnet5large', 'PNASNet-5', '1712.00559', batch_size=BATCH_SIZE//4),
85+
_entry('resnet18', 'ResNet-18', '1812.01187'),
86+
_entry('resnet26', 'ResNet-26', '1812.01187'),
87+
_entry('resnet26d', 'ResNet-26-D', '1812.01187'),
88+
_entry('resnet34', 'ResNet-34', '1812.01187'),
89+
_entry('resnet50', 'ResNet-50', '1812.01187'),
90+
#_entry('resnet101', , ), # same weights as torchvision
91+
#_entry('resnet152', , ), # same weights as torchvision
92+
_entry('resnext50_32x4d', 'ResNeXt-50 32x4d', '1812.01187'),
93+
_entry('resnext50d_32x4d', 'ResNeXt-50-D 32x4d', '1812.01187'),
94+
#_entry('resnext101_32x8d', ), # same weights as torchvision
95+
_entry('semnasnet_100', 'MnasNet-A1', '1807.11626'),
96+
_entry('senet154', 'SENet-154', '1709.01507'),
97+
_entry('seresnet18', 'SE-ResNet-18', '1709.01507'),
98+
_entry('seresnet34', 'SE-ResNet-34', '1709.01507'),
99+
_entry('seresnet50', 'SE-ResNet-50', '1709.01507'),
100+
_entry('seresnet101', 'SE-ResNet-101', '1709.01507'),
101+
_entry('seresnet152', 'SE-ResNet-152', '1709.01507'),
102+
_entry('seresnext26_32x4d', 'SE-ResNeXt-26 32x4d', '1709.01507'),
103+
_entry('seresnext50_32x4d', 'SE-ResNeXt-50 32x4d', '1709.01507'),
104+
_entry('seresnext101_32x4d', 'SE-ResNeXt-101 32x4d', '1709.01507'),
105+
_entry('spnasnet_100', 'Single-Path NAS', '1904.02877'),
106+
_entry('tf_efficientnet_b0', 'EfficientNet-B0 (AutoAugment)', '1905.11946'),
107+
_entry('tf_efficientnet_b1', 'EfficientNet-B1 (AutoAugment)', '1905.11946'),
108+
_entry('tf_efficientnet_b2', 'EfficientNet-B2 (AutoAugment)', '1905.11946'),
109+
_entry('tf_efficientnet_b3', 'EfficientNet-B3 (AutoAugment)', '1905.11946', batch_size=BATCH_SIZE//2),
110+
_entry('tf_efficientnet_b4', 'EfficientNet-B4 (AutoAugment)', '1905.11946', batch_size=BATCH_SIZE//2),
111+
_entry('tf_efficientnet_b5', 'EfficientNet-B5 (AutoAugment)', '1905.11946', batch_size=BATCH_SIZE//4),
112+
_entry('tf_efficientnet_b6', 'EfficientNet-B6 (AutoAugment)', '1905.11946', batch_size=BATCH_SIZE//8),
113+
_entry('tf_efficientnet_b7', 'EfficientNet-B7 (AutoAugment)', '1905.11946', batch_size=BATCH_SIZE//8),
114+
_entry('tf_efficientnet_es', 'EfficientNet-EdgeTPU-S', '1905.11946'),
115+
_entry('tf_efficientnet_em', 'EfficientNet-EdgeTPU-M', '1905.11946'),
116+
_entry('tf_efficientnet_el', 'EfficientNet-EdgeTPU-L', '1905.11946', batch_size=BATCH_SIZE//2),
117+
_entry('tf_inception_v3', 'Inception V3', '1512.00567'),
118+
_entry('tf_mixnet_l', 'MixNet-L', '1907.09595'),
119+
_entry('tf_mixnet_m', 'MixNet-M', '1907.09595'),
120+
_entry('tf_mixnet_s', 'MixNet-S', '1907.09595'),
121+
#_entry('tv_resnet34', , ), # same weights as torchvision
122+
#_entry('tv_resnet50', , ), # same weights as torchvision
123+
#_entry('tv_resnext50_32x4d', , ), # same weights as torchvision
124+
#_entry('wide_resnet50_2' , ), # same weights as torchvision
125+
#_entry('wide_resnet101_2', , ), # same weights as torchvision
126+
_entry('xception', 'Xception', '1610.02357'),
127+
]
195128

196-
for model_name in model_names:
197-
if model_name not in model_map:
198-
print('Skipping %s' % model_name)
199-
continue
200129

130+
for m in model_list:
131+
model_name = m['model']
201132
# create model from name
202133
model = create_model(model_name, pretrained=True)
203134
param_count = sum([m.numel() for m in model.parameters()])
204-
print('Model %s created, param count: %d' % (model_name, param_count))
135+
print('Model %s, %s created. Param count: %d' % (model_name, m['paper_model_name'], param_count))
205136

206137
# get appropriate transform for model's default pretrained config
207-
data_config = resolve_data_config(dict(), model=model, verbose=True)
138+
data_config = resolve_data_config(m['args'], model=model, verbose=True)
139+
if m['ttp']:
140+
model = TestTimePoolHead(model, model.default_cfg['pool_size'])
141+
data_config['crop_pct'] = 1.0
208142
input_transform = create_transform(**data_config)
209143

210144
# Run the benchmark
211145
ImageNet.benchmark(
212146
model=model,
213-
paper_model_name=model_map[model_name]['paper_model_name'],
214-
paper_arxiv_id=model_map[model_name]['paper_arxiv_id'],
147+
paper_model_name=m['paper_model_name'],
148+
paper_arxiv_id=m['paper_arxiv_id'],
215149
input_transform=input_transform,
216-
batch_size=model_map[model_name]['batch_size'],
150+
batch_size=m['batch_size'],
217151
num_gpu=NUM_GPU,
218152
data_root=os.environ.get('IMAGENET_DIR', './imagenet')
219153
)

‎timm/models/helpers.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -60,7 +60,7 @@ def load_pretrained(model, default_cfg, num_classes=1000, in_chans=3, filter_fn=
6060
logging.warning("Pretrained model URL is invalid, using random initialization.")
6161
return
6262

63-
state_dict = model_zoo.load_url(default_cfg['url'])
63+
state_dict = model_zoo.load_url(default_cfg['url'], progress=False)
6464

6565
if in_chans == 1:
6666
conv1_name = default_cfg['first_conv']

0 commit comments

Comments
 (0)