Sample stimuli

How to use

from brainscore_vision import load_benchmark
benchmark = load_benchmark("Geirhos2021colour-error_consistency")
score = benchmark(my_model)

Benchmark API

Code examples

Model scores

Score Legend Min Alignment Max Alignment

Rank	Model	Score
1	SWSL_resnext101_32x8d	.891
2	resnext101_32x8d_wsl	.885
3	convnext_femto_ols:d1_in1k	.866
4	vit_base_patch16_clip_224:openai_ft_in12k_in1k	.815
5	pnasnet_large-tf	.812
6	cvt_cvt-13-384-in22k_finetuned-in1k_4_LucyV4	.811
7	cvt_cvt-13-224-in1k_4_LucyV4	.804
8	pnasnet_large	.800
9	cvt_cvt-21-384-in22k_finetuned-in1k_4_LucyV4	.799
10	vision_transformer_vit_large_patch16_224	.794
11	effnetb1_272x240	.792
12	effnetv2m_custom384	.788
13	nasnet_large-tf	.788
14	resnext101_32x48d_wsl	.787
15	nasnet_large	.769
16	CLIP-RN50	.769
17	SWSL_resnet50	.766
18	vit_large_patch14_clip_224:laion2b_ft_in1k	.762
19	resnext101_32x32d_wsl	.756
20	resnet50-VITO-8deg-cc	.755
21	CLIP_ViT-B_32	.748
22	convnext_tiny:in12k_ft_in1k	.746
23	CLIP_resnet50_float32	.739
24	CLIP_resnet50	.739
25	resnet_50_v2	.738
26	resnet_101_v2	.731
27	cvt_cvt-21-384-in1k_4_LucyV4	.727
28	resnet-152_v2-tf	.721
29	resnet-152_v2_pytorch	.716
30	ViT_L_16_imagenet1k	.715
31	cvt_cvt-21-224-in1k_4_LucyV4	.703
32	resnext101_32x16d_wsl	.696
33	resnet152_imagenet_full	.692
34	resnet_152_v2	.691
35	inception_v3	.689
36	inception_v4	.685
37	efficientnet_b0	.684
38	fixres_resnext101_32x48d_wsl	.675
39	resnet101	.675
40	cvt_cvt-13-384-in1k_4_LucyV4	.670
41	inception_v3-tf	.657
42	resnet101_imagenet_full	.653
43	focalnet_tiny_lrf_in1k	.647
44	ViT_B_16_imagenet1k	.640
45	tv_efficientnet-b1	.637
46	resnet50	.631
47	resnet_152_v1	.624
48	xception	.619
49	resnet50_random_l2_perturb	.617
50	inception_v4-tf	.591
51	resnet-50_v2-tf	.587
52	vit_large_patch14_clip_224:laion2b_ft_in12k_in1k	.587
53	resnet50_imagenet_full	.585
54	antialiased-r50	.574
55	xception-keras	.561
56	vit_relpos_base_patch16_clsgap_224:sw_in1k	.560
57	resnet_101_v1	.551
58	AdvProp_efficientnet-b6	.546
59	effnetb1_cutmixrespatch_SAM_robust32_e10_manylayers_324x288	.543
60	vit_base_patch16_clip_224:openai_ft_in1k	.541
61	effnetb1_cutmixrespatch_SAM_robust32_e2_manylayers_324x288	.538
62	convnext_tiny_imagenet_full_seed-0	.534
63	antialiased-rnext101_32x8d	.528
64	resnet50_byol	.521
65	AdvProp_efficientnet-b7	.521
66	vit_relpos_base_patch32_plus_rpn_256:sw_in1k	.519
67	antialias-resnet152	.512
68	efficientnet-b7	.510
69	vgg_19	.505
70	resnet50_moco_v2	.504
71	cvt_cvt-w24-384-in22k_finetuned-in1k_4	.499
72	Res2Net50_26w_4s	.491
73	effnetb1_cutmixpatch_SAM_robust32_avge6e8e9e10_manylayers_324x288	.491
74	effnetb1_cutmixpatch_augmix_robust32_e5_324x288	.489
75	vit_tiny_r_s16_p8_384:augreg_in21k_ft_in1k	.488
76	densenet-201-keras	.479
77	resnet34_imagenet_full	.478
78	resnet-50-pytorch	.474
79	resnet50-sup	.474
80	resnet50_tutorial	.474
81	resnet_50_v1	.474
82	grcnn_109	.474
83	swin_small_patch4_window7_224:ms_in22k_ft_in1k	.472
84	convnext_large:fb_in22k_ft_in1k	.470
85	AdvProp_efficientnet-b2	.468
86	AT_efficientnet-b2	.468
87	resnet50-SIN_IN_IN	.464
88	resnet_SIN_IN_FT_IN	.464
89	ViT_L_32_imagenet1k	.463
90	convnext_small_imagenet_full_seed-0	.462
91	resnet50-moclr8deg	.461
92	densenet-169	.456
93	deit_small_imagenet_full_seed-0	.454
94	AdvProp_efficientnet-b4	.450
95	AT_efficientnet-b4	.450
96	densenet-169-keras	.448
97	effnetb1_cutmixpatch_augmix_robust32_avge4e7_manylayers_324x288	.448
98	densenet-121-keras	.448
99	cornet_s	.445
100	resnet18_imagenet_full	.443
101	densenet_201_pytorch	.442
102	convnext_xlarge:fb_in22k_ft_in1k	.441
103	grcnn	.439
104	densenet-201	.438
105	resnet-101_v1-tf	.432
106	mobilenet_v2_1_3_224	.430
107	efficientnet_b2_imagenet_full	.429
108	vit_large_patch14_clip_224:openai_ft_in1k	.428
109	resnet50_random_linf8_perturb	.427
110	convnext_base:clip_laiona_augreg_ft_in1k_384	.422
111	resnet50_finetune_cutmix_e3_robust_linf8255_e0_247x234	.419
112	deit_large_imagenet_full_seed-0	.411
113	convnext_small_imagenet_100_seed-0	.406
114	deit_base_imagenet_full_seed-0	.406
115	densenet-121	.404
116	vit_large_patch14_clip_224:openai_ft_in12k_in1k	.403
117	vit_huge_patch14_clip_224:laion2b_ft_in12k_in1k	.403
118	cv_18_dagger_408_pretrained	.400
119	resnet-50_v1-tf	.395
120	convnext_large_imagenet_full_seed-0	.395
121	BiT-S-R50x3	.391
122	convnext_xxlarge:clip_laion2b_soup_ft_in1k	.390
123	vit_large_patch14_clip_336:laion2b_ft_in1k	.387
124	ReAlnet03	.377
125	resnet-152_v1-tf	.373
126	effnetb1_cutmix_augmix_sam_e1_5avg_424x377	.370
127	ReAlnet01	.370
128	ReAlnet_incorrect	.370
129	convnext_base_imagenet_full_seed-0	.365
130	mobilenet_v2_1.4_224-tf	.363
131	voneresnet-50-non_stochastic	.361
132	BiT-S-R152x2	.356
133	custom_model_cv_18_dagger_408	.346
134	resnet50-SIN_IN	.344
135	inception_v2-tf	.344
136	mobilenet_v2_1.3_224-tf	.343
137	ReAlnet04	.342
138	vgg_16	.341
139	ReAlnet08	.328
140	ReAlnet09	.325
141	vit_huge_patch14_clip_336:laion2b_ft_in12k_in1k	.324
142	ReAlnet05	.322
143	efficientnet_b0_imagenet_full	.320
144	ViT_B_32_imagenet1k	.320
145	ReAlnet07	.316
146	CORnet-S	.314
147	ReAlnet06	.314
148	konkle_alexnetgn_ipcl_ref12_supervised_ipcl_aug	.311
149	nasnet_mobile-tf	.309
150	inception_v1	.300
151	ReAlnet02	.299
152	vonegrcnn_62e_nobn	.298
153	vonegrcnn_62e_nobn	.293
154	efficientnet_b1_imagenet_full	.290
155	ReAlnet10	.290
156	mobilenet_v2_1_4_224	.288
157	mobilenet_v2_1-4_224_pytorch	.288
158	inception_v1-tf	.288
159	nasnet_mobile	.286
160	vonegrcnn_47e	.284
161	BiT-S-R50x1	.269
162	resnet50_linf_4_robust	.268
163	AdvProp_efficientnet-b8	.263
164	BiT-S-R101x3	.263
165	BiT-S-R152x4	.261
166	resnet-18	.260
167	vonegrcnn_47e	.260
168	convnext_tiny_sup	.254
169	mobilenet_v2_0_5_224	.253
170	resnet-34	.252
171	vit_large_patch14_clip_336:openai_ft_in12k_in1k	.248
172	regnet_y_400mf	.248
173	mobilenet_v2_1_0_192	.246
174	vonegrcnn_52e_full	.239
175	vonegrcnn_52e_full	.236
176	mobilenet_v2_0_75_224	.231
177	mobilenet_v2_1_0_160	.228
178	voneresnet-50	.218
179	convnext_large_mlp:clip_laion2b_augreg_ft_in1k_384	.216
180	mobilenet_v1_1.0_224	.215
181	blt_vs	.214
182	mobilenet_v2_1_0_224	.214
183	vgg-16-keras	.211
184	mobilenet_v1_0.75_224	.211
185	vgg-19-keras	.188
186	BiT-S-R101x1	.182
187	texture_shape_resnet50_trained_on_SIN	.182
188	voneresnet-50-robust	.180
189	mobilenet_v2_1.0_160-tf	.177
190	mobilenet_v1_1.0_160	.173
191	resnet50-vitoimagevidnet8	.170
192	mobilenet_v2_1.0_224-tf	.168
193	mobilenet_v2_0_5_192	.165
194	alexnet	.163
195	alexnet_random_linf8_perturb	.162
196	mobilenet_v2_1.0_192-tf	.161
197	mobilenet_v2_0.75_224-tf	.159
198	alexnet_random_l2_3_perturb	.152
199	resnet50_robust_l2_eps1	.152
200	mobilenet_v2_0.5_224-tf	.150
201	mobilenet_v2_0_75_192	.150
202	resnet50_l2_3_robust	.146
203	resnet50_simclr	.143
204	alexnet_reduced_aliasing_early_checkpoint	.137
205	mobilenet_v2_0_75_160	.134
206	alexnet	.134
207	alexnet	.134
208	alexnet	.134
209	alexnet	.134
210	alexnet	.134
211	alexnet-baseline	.134
212	alexnet	.134
213	alexnet	.134
214	alexnet	.134
215	alexnet	.134
216	alexnet_ks_torevert	.134
217	konkle_alexnetgn_ipcl_ref01_primary_model	.134
218	mobilenet_v2_0.5_192-tf	.131
219	shufflenet_v2_x1_0	.130
220	mobilenet_v1_0.5_224	.123
221	mobilenet_v2_1_0_128	.122
222	resnet50_imagenet_100_seed-0	.120
223	alexnet_early_checkpoint	.119
224	AdvProp_efficientnet-b0	.115
225	mobilenet_v1_0.75_192	.113
226	resnet50-SIN	.111
227	mobilenet_v2_0.75_160-tf	.108
228	vonealexnet_gaussian_noise_std4_fixed	.104
229	resnet50_robust_l2_eps3	.104
230	imagenet_l2_3_0	.104
231	resnet-50-robust	.104
232	resnet50_linf_8_robust	.104
233	mobilenet_v1_1.0_192	.096
234	mobilenet_v1_0.5_192	.090
235	0.5x_resnet-18_LC_w_sh_1_iter	.084
236	mobilenet_v2_0.75_192-tf	.072
237	mobilenet_v2_0.75_96	.072
238	texture_shape_alexnet_trained_on_SIN	.071
239	mobilenet_v1_0.75_160	.068
240	mobilenet_v1_1.0_128	.068
241	squeezenet1_0	.067
242	AlexNet_SIN	.065
243	mobilenet_v2_0.75_128	.062
244	mobilenet_v2_1.0_128-tf	.061
245	alexnet_linf_8_robust	.060
246	alexnet_l2_3_robust	.060
247	mobilenet_v2_0.35_160	.058
248	alexnet_training_seed_07	.050
249	alexnet_training_seed_04	.049
250	alexnet_training_seed_08	.049
251	alexnet_training_seed_02	.049
252	alexnet_training_seed_10	.049
253	mobilenet_v2_0.35_192	.048
254	squeezenet1_1	.047
255	mobilenet_v1_0.25_192	.045
256	mobilenet_v1_0.25_224	.044
257	alexnet_robust_correct	.043
258	resnet50-vicreg	.043
259	imagenet_l2_10_0	.041
260	mobilenet_v1_0.75_128	.039
261	mobilenet_v2_0.35_128	.038
262	mobilenet_v2_0.5_128	.037
263	tf_efficientnetv2_s_in21ft1k_robust_linf12255_400x400	.037
264	mobilenet_v1_0.5_160	.036
265	mobilenet_v2_0.35_96	.036
266	mobilenet_v2_0.35_224	.035
267	deit_small_imagenet_100_seed-0	.035
268	resnet50-vicregl0p9	.031
269	mobilenet_v1_0.25_128	.030
270	mobilenet_v2_1.0_96	.030
271	convnext_small_imagenet_10_seed-0	.030
272	deit_small_imagenet_10_seed-0	.027
273	resnet50_imagenet_10_seed-0	.027
274	deit_small_imagenet_1_seed-0	.027
275	alexnet_training_seed_03	.025
276	mobilenet_v2_0.5_160	.025
277	mobilenet_v2_0.5_96	.025
278	resnet50-meshes-lt-100-original-pretrained	.022
279	alexnet_training_seed_06	.022
280	resnet50-barlow	.020
281	mobilenet_v1_0.5_128	.020
282	alexnet_training_seed_01	.020
283	alexnet_training_seed_05	.020
284	artResNet18_1	.015
285	my-model	.014
286	resnet50-vicregl0p75	.014
287	barlow-twins-resnet50	.009
288	resnet-50_untrained	.009
289	resnet50-meshes-lt-100-original-scratch	.009
290	convnext_small_imagenet_1_seed-0	.009
291	alexnet_training_seed_09	.007
292	mobilenet_v1_0.25_160	.006
293	resnet50_FractalDB	.004
294	mobilevit_small	.004
295	resnet50_imagenet_1_seed-0	.003
296	resnet50-simclr-vissl	.002
297	briaai_rmbg_1_4	X
298	CORnetZ_CIFAR10_bs32_20_04	X
299	unet_entire	X
300	v1-pyr-nodown	X
301	dcgan	X
302	resnet-50x2_untrained	X
303	resnet-18_untrained	X
304	resnet-18-LC_untrained	X
305	omnivore_swinT	X
306	resnet18-contrastive_predictive	X
307	r3m_resnet34_nocrop	X
308	ViT_B_16	X
309	resnet18-depth_prediction	X
310	yudixie_resnet50_imagenet1kpret_0_240312	X
311	yudixie_resnet18_translation_reg_0_240719	X
312	yolos_tiny	X
313	ReAlnet03_cornet	X
314	ViT_B_32	X
315	yudixie_resnet18_rotation_reg_0_240719	X
316	yudixie_resnet18_imagenet1kpret_0_240719	X
317	ViT_L_32	X
318	ReAlnet02_cornet	X
319	resnet18-simclr	X
320	resnet18-autoencoder	X
321	resnet18_ecoset_full	X
322	RN50	X
323	resnet34_ecoset_full	X
324	resnet50_ecoset_full	X
325	0.5x_resnet-18_LC_w_sh_100_iter	X
326	RN50	X
327	0.5x_resnet-18	X
328	resnet101_ecoset_full	X
329	yudixie_resnet50_distance_rotation_0_240908	X
330	yudixie_resnet50_rotation_reg_0_240908	X
331	resnet152_ecoset_full	X
332	yudixie_resnet18_distance_translation_0_240719	X
333	ReAlnet04_cornet	X
334	yudixie_resnet50_distance_translation_0_240908	X
335	yudixie_resnet18_translation_rotation_0_240719	X
336	yudixie_resnet18_distance_rotation_0_240719	X
337	resnet18-contrastive_multiview	X
338	yudixie_resnet18_category_class_0_240719	X
339	ReAlnet05_cornet	X
340	yudixie_resnet18_random_0_240719	X
341	yudixie_resnet18_cat_obj_class_all_latents_0_240719	X
342	resnet-18_test_m	X
343	yudixie_resnet18_object_class_0_240719	X
344	yudixie_resnet18_distance_translation_rotation_0_240719	X
345	resnet-18-LC_conv_init_m	X
346	yudixie_resnet18_distance_reg_0_240719	X
347	resnet-18-LC_1st_conv_m	X
348	ReAlnet06_cornet	X
349	yudixie_resnet50_translation_reg_0_240908	X
350	resnet18-supervised	X
351	ReAlnet07_cornet	X
352	yudixie_resnet50_translation_rotation_0_240908	X
353	r3m_resnet50_nocrop	X
354	resnet-18-LC_m	X
355	r3m_resnet34	X
356	resnet18-relative_position	X
357	ReAlnet08_cornet	X
358	r3m_resnet50	X
359	VOneCORnet-S	X
360	0.5x_resnet-18_LC_w_sh_10_iter	X
361	omnivore_swinB	X
362	vggface	X
363	ReAlnet09_cornet	X
364	my-model	X
365	CORnet-Z	X
366	resnet18-instance_recognition	X
367	r3m_resnet18	X
368	resnet18-deepcluster	X
369	resnet18-colorization	X
370	CORnetZ_CIFAR10	X
371	0.5x_resnet-18_LC	X
372	omnivore_swinS	X
373	resnet18-local_aggregation	X
374	ReAlnet10_cornet	X
375	FrankRobWobv0	X
376	resnet-18-LC_w_sh_10_iter_conv_init_m	X
377	resnet-18-LC_w_sh_100_iter_m	X
378	resnet-18-LC_w_sh_10_iter_m	X
379	yudixie_resnet50_distance_translation_rotation_0_240908	X
380	resnet-18-LC_w_sh_1_iter_m	X
381	resnet-18-LC_w_sh_1_iter_conv_init_m	X
382	BiT-M-R152x2	X
383	BiT-M-R50x1	X
384	BiT-M-R152x4	X
385	BiT-M-R50x3	X
386	ViT-B/32	X
387	resnet-18-LC_w_sh_100_iter_conv_init_m	X
388	yudixie_resnet50_object_class_0_240908	X
389	yudixie_resnet50_category_class_0_240908	X
390	yudixie_resnet50_cat_obj_class_all_latents_0_240908	X
391	yudixie_resnet50_imagenet1kpret_0_240908	X
392	yudixie_resnet50_random_0_240908	X
393	fulltest_microblockvf_nobottleneck_freeat_nopretrain_eps2_m4	X
394	resnet50_ImageNet	X
395	resnet50_pretrained_with_retinal_waves	X
396	resnet50-cifar	X
397	resnet18_imagenet21kP	X
398	ConvLSTM	X
399	PredRNN	X
400	yudixie_resnet50_distance_reg_0_240908	X
401	SimVP	X
402	TAU	X
403	MIM	X
404	hmax	X
405	openclip	X
406	resnet50	X
407	ReAlnet01_cornet	X
408	pnasnet_large_pytorch	X
409	dorinet_cornet_z	X
410	inception_v3_pytorch	X
411	pixels	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

0.42.

Note that scores are relative to this ceiling.

Data: Geirhos2021colour

Metric: error_consistency