Sample stimuli

How to use

from brainscore_vision import load_benchmark
benchmark = load_benchmark("Geirhos2021powerequalisation-error_consistency")
score = benchmark(my_model)

Benchmark API

Code examples

Model scores

Score Legend Min Alignment Max Alignment

Rank	Model	Score
1	fixres_resnext101_32x48d_wsl	.755
2	vit_large_patch14_clip_224:laion2b_ft_in1k	.718
3	cvt_cvt-13-384-in1k_4_LucyV4	.713
4	vit_large_patch14_clip_224:openai_ft_in1k	.709
5	cvt_cvt-21-384-in1k_4_LucyV4	.707
6	vit_large_patch14_clip_224:openai_ft_in12k_in1k	.707
7	cvt_cvt-w24-384-in22k_finetuned-in1k_4	.705
8	convnext_xlarge:fb_in22k_ft_in1k	.680
9	cvt_cvt-21-384-in22k_finetuned-in1k_4_LucyV4	.672
10	convnext_large:fb_in22k_ft_in1k	.670
11	SWSL_resnext101_32x8d	.668
12	vit_large_patch14_clip_336:laion2b_ft_in1k	.651
13	convnext_xxlarge:clip_laion2b_soup_ft_in1k	.650
14	vit_base_patch16_clip_224:openai_ft_in1k	.647
15	vit_huge_patch14_clip_224:laion2b_ft_in12k_in1k	.638
16	vit_large_patch14_clip_224:laion2b_ft_in12k_in1k	.633
17	resnet_152_v2	.599
18	resnext101_32x16d_wsl	.592
19	convnext_base:clip_laiona_augreg_ft_in1k_384	.586
20	resnext101_32x32d_wsl	.579
21	vit_large_patch14_clip_336:openai_ft_in12k_in1k	.564
22	vit_base_patch16_clip_224:openai_ft_in12k_in1k	.560
23	resnext101_32x48d_wsl	.548
24	convnext_large_mlp:clip_laion2b_augreg_ft_in1k_384	.540
25	effnetv2m_custom384	.501
26	resnet-152_v2_pytorch	.494
27	SWSL_resnet50	.489
28	vit_huge_patch14_clip_336:laion2b_ft_in12k_in1k	.486
29	convnext_base_imagenet_full_seed-0	.482
30	vision_transformer_vit_large_patch16_224	.472
31	convnext_tiny:in12k_ft_in1k	.468
32	convnext_large_imagenet_full_seed-0	.467
33	cvt_cvt-13-384-in22k_finetuned-in1k_4_LucyV4	.462
34	resnet_50_v2	.455
35	resnext101_32x8d_wsl	.453
36	resnet_101_v2	.451
37	convnext_small_imagenet_full_seed-0	.450
38	convnext_tiny_imagenet_full_seed-0	.427
39	CLIP_resnet50_float32	.427
40	CLIP_resnet50	.427
41	cvt_cvt-13-224-in1k_4_LucyV4	.397
42	cv_18_dagger_408_pretrained	.388
43	vit_relpos_base_patch16_clsgap_224:sw_in1k	.382
44	pnasnet_large	.380
45	convnext_femto_ols:d1_in1k	.378
46	nasnet_large	.363
47	ViT_B_16_imagenet1k	.339
48	resnet-152_v2-tf	.335
49	inception_v4	.323
50	focalnet_tiny_lrf_in1k	.317
51	resnet152_imagenet_full	.317
52	CLIP-RN50	.315
53	ViT_L_32_imagenet1k	.307
54	pnasnet_large-tf	.295
55	resnet50-VITO-8deg-cc	.292
56	antialias-resnet152	.289
57	effnetb1_272x240	.289
58	resnet101	.287
59	CLIP_ViT-B_32	.287
60	inception_v3	.280
61	effnetb1_cutmixpatch_SAM_robust32_avge6e8e9e10_manylayers_324x288	.278
62	effnetb1_cutmixrespatch_SAM_robust32_e10_manylayers_324x288	.269
63	deit_large_imagenet_full_seed-0	.269
64	resnet50_random_linf8_perturb	.260
65	resnet_152_v1	.260
66	swin_small_patch4_window7_224:ms_in22k_ft_in1k	.257
67	nasnet_large-tf	.248
68	effnetb1_cutmixrespatch_SAM_robust32_e2_manylayers_324x288	.246
69	xception-keras	.245
70	inception_v3-tf	.237
71	resnet_101_v1	.234
72	cvt_cvt-21-224-in1k_4_LucyV4	.228
73	resnet50_random_l2_perturb	.227
74	inception_v3_pytorch	.222
75	resnet101_imagenet_full	.221
76	effnetb1_cutmixpatch_augmix_robust32_e5_324x288	.217
77	ViT_B_32_imagenet1k	.217
78	resnet50_imagenet_full	.216
79	effnetb1_cutmixpatch_augmix_robust32_avge4e7_manylayers_324x288	.214
80	inception_v4-tf	.209
81	grcnn_109	.201
82	resnet50_finetune_cutmix_e3_robust_linf8255_e0_247x234	.197
83	vgg_19	.195
84	Res2Net50_26w_4s	.193
85	AdvProp_efficientnet-b7	.190
86	AdvProp_efficientnet-b4	.189
87	AT_efficientnet-b4	.189
88	AdvProp_efficientnet-b6	.187
89	AT_efficientnet-b2	.186
90	AdvProp_efficientnet-b2	.186
91	resnet50	.185
92	xception	.184
93	resnet50_linf_4_robust	.182
94	efficientnet_b0	.181
95	vit_relpos_base_patch32_plus_rpn_256:sw_in1k	.173
96	deit_base_imagenet_full_seed-0	.169
97	convnext_small_imagenet_100_seed-0	.167
98	vit_tiny_r_s16_p8_384:augreg_in21k_ft_in1k	.164
99	resnet50-moclr8deg	.159
100	resnet-50_v2-tf	.155
101	antialiased-rnext101_32x8d	.149
102	densenet-121-keras	.147
103	custom_model_cv_18_dagger_408	.145
104	inception_v2-tf	.145
105	alexnet_random_linf8_perturb	.140
106	resnet50_byol	.134
107	pnasnet_large_pytorch	.133
108	resnet_SIN_IN_FT_IN	.132
109	resnet50-SIN_IN_IN	.132
110	BiT-S-R50x3	.128
111	konkle_alexnetgn_ipcl_ref12_supervised_ipcl_aug	.127
112	antialiased-r50	.127
113	resnet34_imagenet_full	.126
114	grcnn	.125
115	densenet_201_pytorch	.125
116	densenet-201	.125
117	voneresnet-50-robust	.122
118	ReAlnet03	.121
119	densenet-121	.119
120	resnet50_linf_8_robust	.119
121	ReAlnet04	.118
122	effnetb1_cutmix_augmix_sam_e1_5avg_424x377	.116
123	ReAlnet08	.115
124	efficientnet-b7	.114
125	vgg_16	.113
126	densenet-201-keras	.112
127	voneresnet-50-non_stochastic	.108
128	resnet-50_v1-tf	.107
129	vonealexnet_gaussian_noise_std4_fixed	.105
130	ReAlnet01	.105
131	ReAlnet_incorrect	.105
132	inception_v1	.100
133	alexnet_l2_3_robust	.098
134	alexnet_random_l2_3_perturb	.098
135	resnet_50_v1	.098
136	resnet50_tutorial	.098
137	resnet50-sup	.098
138	resnet-50-pytorch	.098
139	alexnet_ks_torevert	.097
140	alexnet	.097
141	alexnet	.097
142	alexnet	.097
143	alexnet	.097
144	alexnet	.097
145	alexnet	.097
146	alexnet	.097
147	alexnet	.097
148	alexnet-baseline	.097
149	alexnet	.097
150	ReAlnet05	.096
151	densenet-169	.096
152	resnet50_l2_3_robust	.095
153	resnet-152_v1-tf	.094
154	resnet50_moco_v2	.094
155	densenet-169-keras	.094
156	ReAlnet02	.093
157	ReAlnet07	.091
158	cornet_s	.089
159	convnext_tiny_sup	.089
160	ReAlnet06	.089
161	vonegrcnn_62e_nobn	.088
162	vonegrcnn_62e_nobn	.088
163	alexnet_reduced_aliasing_early_checkpoint	.088
164	nasnet_mobile-tf	.086
165	alexnet_early_checkpoint	.085
166	deit_small_imagenet_full_seed-0	.084
167	BiT-S-R152x2	.083
168	ReAlnet10	.083
169	vonegrcnn_52e_full	.083
170	alexnet_training_seed_05	.083
171	alexnet_training_seed_01	.083
172	resnet50-vitoimagevidnet8	.082
173	alexnet	.082
174	inception_v1-tf	.078
175	barlow-twins-resnet50	.078
176	ReAlnet09	.078
177	resnet-101_v1-tf	.076
178	resnet50_robust_l2_eps1	.075
179	vonegrcnn_47e	.075
180	AdvProp_efficientnet-b8	.073
181	resnet-34	.070
182	vonegrcnn_52e_full	.070
183	squeezenet1_1	.068
184	resnet18_imagenet_full	.065
185	vonegrcnn_47e	.065
186	mobilenet_v2_1.0_160-tf	.065
187	alexnet_robust_correct	.064
188	artResNet18_1	.064
189	resnet50-SIN_IN	.063
190	resnet-50-robust	.062
191	imagenet_l2_3_0	.062
192	resnet50_robust_l2_eps3	.062
193	konkle_alexnetgn_ipcl_ref01_primary_model	.062
194	mobilenet_v2_1_0_192	.060
195	resnet50_simclr	.059
196	mobilenet_v2_1_0_160	.059
197	BiT-S-R152x4	.057
198	alexnet_training_seed_04	.053
199	alexnet_training_seed_08	.053
200	alexnet_training_seed_02	.053
201	alexnet_training_seed_10	.053
202	resnet50-meshes-lt-100-original-scratch	.053
203	vgg-19-keras	.053
204	CORnet-S	.052
205	efficientnet_b2_imagenet_full	.050
206	mobilenet_v1_0.25_128	.048
207	mobilenet_v2_1.0_224-tf	.047
208	vgg-16-keras	.047
209	mobilenet_v2_0_5_192	.046
210	voneresnet-50	.045
211	texture_shape_resnet50_trained_on_SIN	.044
212	mobilenet_v2_0.5_96	.043
213	mobilenet_v1_0.5_160	.041
214	convnext_small_imagenet_1_seed-0	.041
215	deit_small_imagenet_100_seed-0	.040
216	mobilenet_v2_1_0_224	.040
217	resnet50_imagenet_100_seed-0	.040
218	resnet50-SIN	.040
219	mobilenet_v1_0.5_128	.039
220	BiT-S-R101x3	.039
221	resnet50-barlow	.038
222	alexnet_training_seed_07	.038
223	alexnet_training_seed_09	.038
224	AdvProp_efficientnet-b0	.038
225	0.5x_resnet-18_LC_w_sh_1_iter	.038
226	BiT-S-R50x1	.037
227	mobilenet_v2_0.35_192	.037
228	texture_shape_alexnet_trained_on_SIN	.037
229	alexnet_linf_8_robust	.037
230	mobilenet_v2_1_4_224	.036
231	mobilenet_v2_1-4_224_pytorch	.036
232	resnet50_FractalDB	.036
233	BiT-S-R101x1	.035
234	mobilenet_v1_0.75_192	.035
235	mobilenet_v1_0.5_224	.034
236	mobilenet_v2_0.5_128	.034
237	mobilenet_v2_0_75_224	.034
238	mobilenet_v1_0.25_192	.033
239	mobilenet_v2_0.35_224	.033
240	resnet50-meshes-lt-100-original-pretrained	.033
241	resnet50_imagenet_10_seed-0	.033
242	AlexNet_SIN	.032
243	nasnet_mobile	.032
244	efficientnet_b0_imagenet_full	.031
245	mobilenet_v2_1.0_96	.031
246	deit_small_imagenet_10_seed-0	.031
247	deit_small_imagenet_1_seed-0	.031
248	convnext_small_imagenet_10_seed-0	.031
249	resnet50-vicregl0p9	.030
250	resnet50-vicregl0p75	.029
251	mobilenet_v2_0.5_192-tf	.028
252	resnet50-vicreg	.027
253	mobilenet_v2_0.75_96	.027
254	mobilenet_v2_1_3_224	.027
255	mobilenet_v2_1.0_128-tf	.026
256	resnet-50_untrained	.024
257	mobilenet_v1_0.25_224	.024
258	mobilenet_v2_0.75_224-tf	.023
259	blt_vs	.023
260	mobilenet_v2_0.35_128	.023
261	mobilenet_v1_0.5_192	.021
262	tf_efficientnetv2_s_in21ft1k_robust_linf12255_400x400	.021
263	mobilenet_v2_1.4_224-tf	.020
264	mobilenet_v2_0.35_160	.020
265	mobilenet_v2_0_75_160	.019
266	mobilenet_v2_0.75_128	.019
267	mobilenet_v2_0_75_192	.019
268	mobilenet_v2_0.5_160	.018
269	imagenet_l2_10_0	.018
270	mobilenet_v2_1.0_192-tf	.017
271	mobilenet_v1_0.75_160	.016
272	squeezenet1_0	.015
273	shufflenet_v2_x1_0	.015
274	mobilenet_v2_0.35_96	.014
275	alexnet_training_seed_06	.014
276	mobilenet_v2_0.5_224-tf	.014
277	mobilenet_v1_0.75_128	.014
278	resnet50_imagenet_1_seed-0	.012
279	mobilenet_v2_0.75_160-tf	.011
280	alexnet_training_seed_03	.010
281	mobilenet_v2_0_5_224	.010
282	mobilevit_small	.010
283	mobilenet_v1_1.0_192	.009
284	resnet50-simclr-vissl	.008
285	mobilenet_v1_0.25_160	.007
286	mobilenet_v1_1.0_128	.006
287	mobilenet_v1_1.0_160	.006
288	mobilenet_v1_1.0_224	.005
289	resnet-18	.004
290	mobilenet_v2_1_0_128	.003
291	mobilenet_v2_1.3_224-tf	.003
292	efficientnet_b1_imagenet_full	.003
293	regnet_y_400mf	.002
294	mobilenet_v2_0.75_192-tf	.002
295	mobilenet_v1_0.75_224	.001
296	my-model	.000
297	yudixie_resnet18_distance_reg_0_240719	X
298	dorinet_cornet_z	X
299	CORnetZ_CIFAR10_bs32_20_04	X
300	resnet50	X
301	openclip	X
302	v1-pyr-nodown	X
303	unet_entire	X
304	dcgan	X
305	resnet-50x2_untrained	X
306	resnet-18_untrained	X
307	resnet-18-LC_untrained	X
308	omnivore_swinT	X
309	resnet18-contrastive_predictive	X
310	r3m_resnet34_nocrop	X
311	ViT_B_16	X
312	resnet18-depth_prediction	X
313	ViT_B_32	X
314	ViT_L_32	X
315	resnet18-simclr	X
316	resnet18-autoencoder	X
317	resnet18-relative_position	X
318	RN50	X
319	0.5x_resnet-18_LC_w_sh_100_iter	X
320	RN50	X
321	resnet50-cifar	X
322	resnet50_pretrained_with_retinal_waves	X
323	resnet50_ImageNet	X
324	fulltest_microblockvf_nobottleneck_freeat_nopretrain_eps2_m4	X
325	0.5x_resnet-18	X
326	resnet18-contrastive_multiview	X
327	resnet-18-LC_conv_init_m	X
328	resnet-18-LC_1st_conv_m	X
329	resnet18-supervised	X
330	r3m_resnet50_nocrop	X
331	resnet-18-LC_m	X
332	resnet18-colorization	X
333	yudixie_resnet50_random_0_240908	X
334	r3m_resnet34	X
335	r3m_resnet50	X
336	yudixie_resnet50_cat_obj_class_all_latents_0_240908	X
337	yudixie_resnet50_imagenet1kpret_0_240908	X
338	yudixie_resnet50_imagenet1kpret_0_240312	X
339	yudixie_resnet50_object_class_0_240908	X
340	0.5x_resnet-18_LC_w_sh_10_iter	X
341	my-model	X
342	vggface	X
343	resnet18-instance_recognition	X
344	CORnet-Z	X
345	yudixie_resnet50_category_class_0_240908	X
346	r3m_resnet18	X
347	resnet18-deepcluster	X
348	resnet18_imagenet21kP	X
349	yudixie_resnet50_distance_reg_0_240908	X
350	ConvLSTM	X
351	PredRNN	X
352	SimVP	X
353	TAU	X
354	MIM	X
355	hmax	X
356	resnet18-local_aggregation	X
357	CORnetZ_CIFAR10	X
358	0.5x_resnet-18_LC	X
359	ReAlnet01_cornet	X
360	FrankRobWobv0	X
361	yudixie_resnet18_translation_reg_0_240719	X
362	ReAlnet03_cornet	X
363	pixels	X
364	yudixie_resnet18_rotation_reg_0_240719	X
365	yudixie_resnet18_imagenet1kpret_0_240719	X
366	ReAlnet02_cornet	X
367	resnet-18-LC_w_sh_1_iter_m	X
368	resnet-18-LC_w_sh_10_iter_conv_init_m	X
369	resnet18_ecoset_full	X
370	resnet-18-LC_w_sh_100_iter_m	X
371	resnet34_ecoset_full	X
372	resnet50_ecoset_full	X
373	VOneCORnet-S	X
374	resnet101_ecoset_full	X
375	yudixie_resnet50_distance_translation_0_240908	X
376	ReAlnet04_cornet	X
377	resnet152_ecoset_full	X
378	yudixie_resnet18_distance_translation_0_240719	X
379	resnet-18-LC_w_sh_10_iter_m	X
380	yudixie_resnet18_translation_rotation_0_240719	X
381	yudixie_resnet18_distance_rotation_0_240719	X
382	yudixie_resnet18_category_class_0_240719	X
383	resnet-18_test_m	X
384	yudixie_resnet18_random_0_240719	X
385	yudixie_resnet18_cat_obj_class_all_latents_0_240719	X
386	yudixie_resnet50_rotation_reg_0_240908	X
387	yudixie_resnet50_distance_rotation_0_240908	X
388	yudixie_resnet18_distance_translation_rotation_0_240719	X
389	yudixie_resnet18_object_class_0_240719	X
390	ReAlnet05_cornet	X
391	ReAlnet06_cornet	X
392	yudixie_resnet50_translation_reg_0_240908	X
393	yudixie_resnet50_translation_rotation_0_240908	X
394	ReAlnet07_cornet	X
395	ReAlnet08_cornet	X
396	resnet-18-LC_w_sh_1_iter_conv_init_m	X
397	BiT-M-R152x2	X
398	omnivore_swinB	X
399	ReAlnet09_cornet	X
400	omnivore_swinS	X
401	yudixie_resnet50_distance_translation_rotation_0_240908	X
402	ReAlnet10_cornet	X
403	BiT-M-R50x1	X
404	BiT-M-R50x3	X
405	ViT-B/32	X
406	resnet-18-LC_w_sh_100_iter_conv_init_m	X

Benchmark bibtex

@article{geirhos2021partial,
              title={Partial success in closing the gap between human and machine vision},
              author={Geirhos, Robert and Narayanappa, Kantharaju and Mitzkus, Benjamin and Thieringer, Tizian and Bethge, Matthias and Wichmann, Felix A and Brendel, Wieland},
              journal={Advances in Neural Information Processing Systems},
              volume={34},
              year={2021},
              url={https://openreview.net/forum?id=QkljT4mrfs}
        }

Ceiling

0.51.

Note that scores are relative to this ceiling.

Data: Geirhos2021powerequalisation

Metric: error_consistency