Leaderboard Explore Compare Profile Community Blog FAQ Competition 2024 Report Issue

vit_large_patch14_clip_224:openai_ft_in1k

Contributed by Anonymous User No reference provided Find on GitHub source Compare this model

Vision transformer

≈304.2M parameters

224 × 224 input

Contrastive + supervised Apache-2.0

Scores on benchmarks

from scoring API

Model rank shown below is with respect to all public models.

.477	average_vision rank 1 138 benchmarks	.477 0 ceiling best median

.351	neural_vision rank 6 77 benchmarks	.351 0 ceiling best median

.448	V1 rank 8 33 benchmarks	.448 0 ceiling best median

.421	Allen2022_fmri_surface.V1 rank 5 2 benchmarks	.421 0 ceiling best median

.302	Allen2022_fmri_surface.V1-rdm v1 [reference] rank 19	.302 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.541	Allen2022_fmri_surface.V1-ridge v1 [reference] rank 4	.541 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.227	FreemanZiemba2013.V1-pls v3 [reference] rank 290	.227 0 ceiling best median
	recordings from 102 sites in V1 315 images Sample stimuli cannot be displayed publicly for this benchmark.

.636	Hebart2023_fmri.V1-ridgecv v3 rank 9	.636 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.236	Li2026.V1 rank 87 2 benchmarks	.236 0 ceiling best median

.126	Li2026.V1-rdm v1 [reference] rank 88	.126 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.345	Li2026.V1-ridgecv v1 [reference] rank 92	.345 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.676	Marques2020 [reference] rank 325 22 benchmarks	.676 0 ceiling best median

.744	V1-orientation rank 367 7 benchmarks	.744 0 ceiling best median

.983	Marques2020_DeValois1982-pref_or v1 rank 48	.983 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.621	Marques2020_Ringach2002-circular_variance v1 rank 397	.621 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.856	Marques2020_Ringach2002-cv_bandwidth_ratio v1 rank 197	.856 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.720	Marques2020_Ringach2002-opr_cv_diff v1 rank 385	.720 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.762	Marques2020_Ringach2002-or_bandwidth v1 rank 329	.762 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.676	Marques2020_Ringach2002-or_selective v1 rank 408	.676 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.589	Marques2020_Ringach2002-orth_pref_ratio v1 rank 386	.589 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.451	V1-receptive_field_size rank 342 2 benchmarks	.451 0 ceiling best median

.370	Marques2020_Cavanaugh2002-grating_summation_field v1 [reference] rank 376	.370 0 ceiling best median
	2304 images Sample stimuli cannot be displayed publicly for this benchmark.

.532	Marques2020_Cavanaugh2002-surround_diameter v1 [reference] rank 202	.532 0 ceiling best median
	2304 images Sample stimuli cannot be displayed publicly for this benchmark.

.928	V1-response_magnitude rank 81 3 benchmarks	.928 0 ceiling best median

.894	Marques2020_FreemanZiemba2013-max_noise v1 [reference] rank 89	.894 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.905	Marques2020_FreemanZiemba2013-max_texture v1 [reference] rank 153	.905 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.985	Marques2020_Ringach2002-max_dc v1 rank 47	.985 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.754	V1-response_selectivity rank 38 4 benchmarks	.754 0 ceiling best median

.818	Marques2020_FreemanZiemba2013-texture_selectivity v1 [reference] rank 108	.818 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.832	Marques2020_FreemanZiemba2013-texture_sparseness v1 [reference] rank 60	.832 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.812	Marques2020_FreemanZiemba2013-texture_variance_ratio v1 [reference] rank 119	.812 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.556	Marques2020_Ringach2002-modulation_ratio v1 rank 141	.556 0 ceiling best median
	1152 images Sample stimuli cannot be displayed publicly for this benchmark.

.609	V1-spatial_frequency rank 384 3 benchmarks	.609 0 ceiling best median

.652	Marques2020_DeValois1982-peak_sf v1 rank 265	.652 0 ceiling best median
	2112 images Sample stimuli cannot be displayed publicly for this benchmark.

.773	Marques2020_Schiller1976-sf_bandwidth v1 [reference] rank 296	.773 0 ceiling best median
	2112 images Sample stimuli cannot be displayed publicly for this benchmark.

.404	Marques2020_Schiller1976-sf_selective v1 [reference] rank 417	.404 0 ceiling best median
	2112 images Sample stimuli cannot be displayed publicly for this benchmark.

.603	V1-surround_modulation rank 252 1 benchmark	.603 0 ceiling best median

.603	Marques2020_Cavanaugh2002-surround_suppression_index v1 [reference] rank 252	.603 0 ceiling best median
	2304 images Sample stimuli cannot be displayed publicly for this benchmark.

.639	V1-texture_modulation rank 193 2 benchmarks	.639 0 ceiling best median

.525	Marques2020_FreemanZiemba2013-abs_texture_modulation_index v1 [reference] rank 216	.525 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.752	Marques2020_FreemanZiemba2013-texture_modulation_index v1 [reference] rank 141	.752 0 ceiling best median
	450 images Sample stimuli cannot be displayed publicly for this benchmark.

.809	Papale2025.V1-ridgecv v3 [reference] rank 10	.809 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.176	Coggan2024_fMRI.V1-rdm v1 rank 50	.176 0 ceiling best median
	24 images Sample stimuli cannot be displayed publicly for this benchmark.

.400	Zerbe2026_fmri.V1 [reference] rank 2 3 benchmarks	.400 0 ceiling best median

.394	Zerbe2026_fmri.V1-ood-ridgecv v1 [reference] rank 12	.394 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.319	Zerbe2026_fmri.V1-rdm-pearson v1 [reference] rank 6	.319 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.488	Zerbe2026_fmri.V1-tau-ridgecv v1 [reference] rank 5	.488 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.286	V2 rank 90 10 benchmarks	.286 0 ceiling best median

.361	Allen2022_fmri_surface.V2 rank 55 2 benchmarks	.361 0 ceiling best median

.260	Allen2022_fmri_surface.V2-rdm v1 [reference] rank 63	.260 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.461	Allen2022_fmri_surface.V2-ridge v1 [reference] rank 75	.461 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.251	FreemanZiemba2013.V2-pls v3 [reference] rank 228	.251 0 ceiling best median
	recordings from 103 sites in V2 315 images Sample stimuli cannot be displayed publicly for this benchmark.

.579	Hebart2023_fmri.V2-ridgecv v3 rank 81	.579 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.141	Li2026.V2 rank 121 2 benchmarks	.141 0 ceiling best median

.053	Li2026.V2-rdm v1 [reference] rank 89	.053 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.228	Li2026.V2-ridgecv v1 [reference] rank 111	.228 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.048	Coggan2024_fMRI.V2-rdm v1 rank 190	.048 0 ceiling best median
	24 images Sample stimuli cannot be displayed publicly for this benchmark.

.338	Zerbe2026_fmri.V2 [reference] rank 38 3 benchmarks	.338 0 ceiling best median

.367	Zerbe2026_fmri.V2-ood-ridgecv v1 [reference] rank 44	.367 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.236	Zerbe2026_fmri.V2-rdm-pearson v1 [reference] rank 60	.236 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.412	Zerbe2026_fmri.V2-tau-ridgecv v1 [reference] rank 42	.412 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.328	V4 rank 25 16 benchmarks	.328 0 ceiling best median

.331	Allen2022_fmri_surface.V4 rank 37 2 benchmarks	.331 0 ceiling best median

.225	Allen2022_fmri_surface.V4-rdm v1 [reference] rank 53	.225 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.438	Allen2022_fmri_surface.V4-ridge v1 [reference] rank 28	.438 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.295	Cowley2026.V4-pls v1 [reference] rank 64	.295 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.324	Hebart2023_fmri.V4-ridgecv v3 rank 55	.324 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.113	Li2026.V4 rank 81 2 benchmarks	.113 0 ceiling best median

.068	Li2026.V4-rdm v1 [reference] rank 91	.068 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.157	Li2026.V4-ridgecv v1 [reference] rank 58	.157 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.029	MajajHong2015public.V4-reverse_pls v4 [reference] rank 44	.029 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.510	MajajHong2015.V4-pls v4 [reference] rank 249	.510 0 ceiling best median
	recordings from 88 sites in V4 2560 images Sample stimuli cannot be displayed publicly for this benchmark.

.593	Papale2025.V4-ridgecv v3 [reference] rank 29	.593 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.545	Sanghavi2020.V4-pls v2 [reference] rank 271	.545 0 ceiling best median
	recordings from 47 sites in V4 5760 images Sample stimuli cannot be displayed publicly for this benchmark.

.448	SanghaviJozwik2020.V4-pls v2 [reference] rank 252	.448 0 ceiling best median
	recordings from 50 sites in V4 4916 images Sample stimuli cannot be displayed publicly for this benchmark.

.231	SanghaviMurty2020.V4-pls v2 [reference] rank 71	.231 0 ceiling best median
	recordings from 46 sites in V4 300 images Sample stimuli cannot be displayed publicly for this benchmark.

.197	Coggan2024_fMRI.V4-rdm v1 rank 26	.197 0 ceiling best median
	24 images Sample stimuli cannot be displayed publicly for this benchmark.

.320	Zerbe2026_fmri.V4 [reference] rank 25 3 benchmarks	.320 0 ceiling best median

.392	Zerbe2026_fmri.V4-ood-ridgecv v1 [reference] rank 53	.392 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.200	Zerbe2026_fmri.V4-rdm-pearson v1 [reference] rank 44	.200 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.369	Zerbe2026_fmri.V4-tau-ridgecv v1 [reference] rank 10	.369 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.343	IT rank 23 18 benchmarks	.343 0 ceiling best median

.427	Allen2022_fmri_surface.IT rank 32 2 benchmarks	.427 0 ceiling best median

.347	Allen2022_fmri_surface.IT-rdm v1 [reference] rank 40	.347 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.507	Allen2022_fmri_surface.IT-ridge v1 [reference] rank 49	.507 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.251	Bracci2019.anteriorVTC-rdm v1 rank 152	.251 0 ceiling best median
	27 images Sample stimuli cannot be displayed publicly for this benchmark.

.250	Gifford2022.IT-ridgecv v3 [reference] rank 31	.250 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.301	Hebart2023_fmri.IT-ridgecv v3 rank 56	.301 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.310	Li2026.IT rank 59 2 benchmarks	.310 0 ceiling best median

.381	Li2026.IT-rdm v1 [reference] rank 62	.381 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.239	Li2026.IT-ridgecv v1 [reference] rank 71	.239 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.399	MajajHong2015.IT-pls v4 [reference] rank 311	.399 0 ceiling best median
	recordings from 168 sites in IT 2560 images Sample stimuli cannot be displayed publicly for this benchmark.

.084	MajajHong2015public.IT-reverse_pls v4 [reference] rank 38	.084 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.584	Papale2025.IT-ridgecv v3 [reference] rank 21	.584 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.455	Sanghavi2020.IT-pls v2 [reference] rank 185	.455 0 ceiling best median
	recordings from 88 sites in IT 5760 images Sample stimuli cannot be displayed publicly for this benchmark.

.399	SanghaviJozwik2020.IT-pls v2 [reference] rank 289	.399 0 ceiling best median
	recordings from 26 sites in IT 4916 images Sample stimuli cannot be displayed publicly for this benchmark.

.337	SanghaviMurty2020.IT-pls v2 [reference] rank 112	.337 0 ceiling best median
	recordings from 29 sites in IT 300 images Sample stimuli cannot be displayed publicly for this benchmark.

.697	Coggan2024_fMRI.IT-rdm v1 rank 27	.697 0 ceiling best median
	24 images Sample stimuli cannot be displayed publicly for this benchmark.

.307	Zerbe2026_fmri.IT [reference] rank 25 3 benchmarks	.307 0 ceiling best median

.220	Zerbe2026_fmri.IT-ood-ridgecv v1 [reference] rank 53	.220 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.406	Zerbe2026_fmri.IT-rdm-pearson v1 [reference] rank 12	.406 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.294	Zerbe2026_fmri.IT-tau-ridgecv v1 [reference] rank 69	.294 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.604	behavior_vision rank 1 61 benchmarks	.604 0 ceiling best median

.534	Rajalingham2018-i2n v2 [reference] rank 104	.534 0 ceiling best median
	match-to-sample task 240 images Sample stimuli cannot be displayed publicly for this benchmark.

.627	Geirhos2021-error_consistency [reference] rank 8 17 benchmarks	.627 0 ceiling best median

.428	Geirhos2021colour-error_consistency v1 [reference] rank 123	.428 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.241	Geirhos2021contrast-error_consistency v1 [reference] rank 119	.241 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.840	Geirhos2021cueconflict-error_consistency v1 [reference] rank 10	.840 0 ceiling best median
	1280 images Sample stimuli cannot be displayed publicly for this benchmark.

.807	Geirhos2021edge-error_consistency v1 [reference] rank 15	.807 0 ceiling best median
	160 images Sample stimuli cannot be displayed publicly for this benchmark.

.619	Geirhos2021eidolonI-error_consistency v1 [reference] rank 26	.619 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.603	Geirhos2021eidolonII-error_consistency v1 [reference] rank 38	.603 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.438	Geirhos2021eidolonIII-error_consistency v1 [reference] rank 81	.438 0 ceiling best median
	480 images Sample stimuli cannot be displayed publicly for this benchmark.

.337	Geirhos2021falsecolour-error_consistency v1 [reference] rank 136	.337 0 ceiling best median
	560 images Sample stimuli cannot be displayed publicly for this benchmark.

.807	Geirhos2021highpass-error_consistency v1 [reference] rank 6	.807 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.663	Geirhos2021lowpass-error_consistency v1 [reference] rank 6	.663 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.576	Geirhos2021phasescrambling-error_consistency v1 [reference] rank 18	.576 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.709	Geirhos2021powerequalisation-error_consistency v1 [reference] rank 6	.709 0 ceiling best median
	560 images Sample stimuli cannot be displayed publicly for this benchmark.

.488	Geirhos2021rotation-error_consistency v1 [reference] rank 27	.488 0 ceiling best median
	960 images Sample stimuli cannot be displayed publicly for this benchmark.

1.0	Geirhos2021silhouette-error_consistency v1 [reference] rank 1	1.0 0 ceiling best median
	160 images Sample stimuli cannot be displayed publicly for this benchmark.

.776	Geirhos2021sketch-error_consistency v1 [reference] rank 2	.776 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.697	Geirhos2021stylized-error_consistency v1 [reference] rank 29	.697 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.639	Geirhos2021uniformnoise-error_consistency v1 [reference] rank 22	.639 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.648	Baker2022 rank 32 3 benchmarks	.648 0 ceiling best median

.858	Baker2022fragmented-accuracy_delta v1 [reference] rank 46	.858 0 ceiling best median
	716 images Sample stimuli cannot be displayed publicly for this benchmark.

.899	Baker2022frankenstein-accuracy_delta v1 [reference] rank 10	.899 0 ceiling best median
	716 images Sample stimuli cannot be displayed publicly for this benchmark.

.186	Baker2022inverted-accuracy_delta v1 [reference] rank 73	.186 0 ceiling best median
	360 images Sample stimuli cannot be displayed publicly for this benchmark.

.812	BMD2024 rank 5 4 benchmarks	.812 0 ceiling best median

.801	BMD2024.dotted_1Behavioral-accuracy_distance v1 rank 5	.801 0 ceiling best median
	100 images Sample stimuli cannot be displayed publicly for this benchmark.

.626	BMD2024.dotted_2Behavioral-accuracy_distance v1 rank 3	.626 0 ceiling best median
	100 images Sample stimuli cannot be displayed publicly for this benchmark.

.941	BMD2024.texture_1Behavioral-accuracy_distance v1 rank 6	.941 0 ceiling best median
	100 images Sample stimuli cannot be displayed publicly for this benchmark.

.882	BMD2024.texture_2Behavioral-accuracy_distance v1 rank 8	.882 0 ceiling best median
	100 images Sample stimuli cannot be displayed publicly for this benchmark.

.633	Ferguson2024 [reference] rank 25 14 benchmarks	.633 0 ceiling best median

.320	Ferguson2024circle_line-value_delta v1 [reference] rank 152	.320 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

1.0	Ferguson2024color-value_delta v1 [reference] rank 1	1.0 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.839	Ferguson2024convergence-value_delta v1 [reference] rank 41	.839 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.361	Ferguson2024eighth-value_delta v1 [reference] rank 81	.361 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.845	Ferguson2024gray_easy-value_delta v1 [reference] rank 30	.845 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.942	Ferguson2024gray_hard-value_delta v1 [reference] rank 28	.942 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.771	Ferguson2024half-value_delta v1 [reference] rank 73	.771 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.020	Ferguson2024juncture-value_delta v1 [reference] rank 263	.020 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.276	Ferguson2024lle-value_delta v1 [reference] rank 207	.276 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

1.0	Ferguson2024llh-value_delta v1 [reference] rank 1	1.0 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.379	Ferguson2024quarter-value_delta v1 [reference] rank 150	.379 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.638	Ferguson2024round_f-value_delta v1 [reference] rank 79	.638 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.820	Ferguson2024round_v-value_delta v1 [reference] rank 66	.820 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.657	Ferguson2024tilted_line-value_delta v1 [reference] rank 124	.657 0 ceiling best median
	2_way_afc task 48 images Sample stimuli cannot be displayed publicly for this benchmark.

.157	Hebart2023-match v1 rank 225	.157 0 ceiling best median
	1854 images Sample stimuli cannot be displayed publicly for this benchmark.

.711	Kato2026 rank 7 18 benchmarks	.711 0 ceiling best median

.618	Kato2026-error_consistency rank 6 9 benchmarks	.618 0 ceiling best median

.630	Kato2026Ori-error_consistency v1 rank 9	.630 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.502	Kato2026Fil-error_consistency v1 rank 87	.502 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.630	Kato2026Lin-error_consistency v1 rank 27	.630 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.598	Kato2026SilTex-error_consistency v1 rank 24	.598 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.746	Kato2026Sil-error_consistency v1 rank 19	.746 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.932	Kato2026Tex-error_consistency v1 rank 12	.932 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.780	Kato2026LinFil-error_consistency v1 rank 8	.780 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.442	Kato2026Sparse-error_consistency v1 rank 7	.442 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.303	Kato2026Dense-error_consistency v1 rank 17	.303 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.803	Kato2026-accuracy_distance rank 12 9 benchmarks	.803 0 ceiling best median

1.0	Kato2026Ori-accuracy_distance v1 rank 1	1.0 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.755	Kato2026Fil-accuracy_distance v1 rank 135	.755 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.945	Kato2026Lin-accuracy_distance v1 rank 16	.945 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.826	Kato2026SilTex-accuracy_distance v1 rank 26	.826 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.707	Kato2026Sil-accuracy_distance v1 rank 21	.707 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

1.0	Kato2026Tex-accuracy_distance v1 rank 1	1.0 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

1.0	Kato2026LinFil-accuracy_distance v1 rank 1	1.0 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.526	Kato2026Sparse-accuracy_distance v1 rank 10	.526 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.471	Kato2026Dense-accuracy_distance v1 rank 23	.471 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.784	Maniquet2024 rank 22 2 benchmarks	.784 0 ceiling best median

.826	Maniquet2024-confusion_similarity v1 [reference] rank 39	.826 0 ceiling best median
	13600 images Sample stimuli cannot be displayed publicly for this benchmark.

.742	Maniquet2024-tasks_consistency v1 [reference] rank 13	.742 0 ceiling best median
	13600 images Sample stimuli cannot be displayed publicly for this benchmark.

.526	Coggan2024_behavior-ConditionWiseAccuracySimilarity v1 rank 51	.526 0 ceiling best median
	22560 images Sample stimuli cannot be displayed publicly for this benchmark.

.554	engineering_vision rank 9 25 benchmarks	.554 0 ceiling best median

.850	ImageNet-top1 v1 [reference] rank 12	.850 0 ceiling best median
	50000 images Sample stimuli cannot be displayed publicly for this benchmark.

.723	ImageNet-C-top1 [reference] rank 3 4 benchmarks	.723 0 ceiling best median

.734	ImageNet-C-noise-top1 v2 [reference] rank 3	.734 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.650	ImageNet-C-blur-top1 v2 [reference] rank 3	.650 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.762	ImageNet-C-weather-top1 v2 [reference] rank 1	.762 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.747	ImageNet-C-digital-top1 v2 [reference] rank 6	.747 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.809	Geirhos2021-top1 [reference] rank 10 17 benchmarks	.809 0 ceiling best median

.995	Geirhos2021colour-top1 v1 [reference] rank 22	.995 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.995	Geirhos2021contrast-top1 v1 [reference] rank 10	.995 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.377	Geirhos2021cueconflict-top1 v1 [reference] rank 35	.377 0 ceiling best median
	1280 images Sample stimuli cannot be displayed publicly for this benchmark.

.894	Geirhos2021edge-top1 v1 [reference] rank 13	.894 0 ceiling best median
	160 images Sample stimuli cannot be displayed publicly for this benchmark.

.606	Geirhos2021eidolonI-top1 v1 [reference] rank 15	.606 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.638	Geirhos2021eidolonII-top1 v1 [reference] rank 19	.638 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.708	Geirhos2021eidolonIII-top1 v1 [reference] rank 17	.708 0 ceiling best median
	480 images Sample stimuli cannot be displayed publicly for this benchmark.

.995	Geirhos2021falsecolour-top1 v1 [reference] rank 13	.995 0 ceiling best median
	560 images Sample stimuli cannot be displayed publicly for this benchmark.

.955	Geirhos2021highpass-top1 v1 [reference] rank 13	.955 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.641	Geirhos2021lowpass-top1 v1 [reference] rank 13	.641 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.892	Geirhos2021phasescrambling-top1 v1 [reference] rank 10	.892 0 ceiling best median
	640 images Sample stimuli cannot be displayed publicly for this benchmark.

.964	Geirhos2021powerequalisation-top1 v1 [reference] rank 12	.964 0 ceiling best median
	560 images Sample stimuli cannot be displayed publicly for this benchmark.

.964	Geirhos2021rotation-top1 v1 [reference] rank 14	.964 0 ceiling best median
	960 images Sample stimuli cannot be displayed publicly for this benchmark.

.719	Geirhos2021silhouette-top1 v1 [reference] rank 7	.719 0 ceiling best median
	160 images Sample stimuli cannot be displayed publicly for this benchmark.

.958	Geirhos2021sketch-top1 v1 [reference] rank 3	.958 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.638	Geirhos2021stylized-top1 v1 [reference] rank 26	.638 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.821	Geirhos2021uniformnoise-top1 v1 [reference] rank 6	.821 0 ceiling best median
	800 images Sample stimuli cannot be displayed publicly for this benchmark.

.386	Hermann2020 [reference] rank 43 2 benchmarks	.386 0 ceiling best median

.432	Hermann2020cueconflict-shape_bias v1 [reference] rank 69	.432 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

.339	Hermann2020cueconflict-shape_match v1 [reference] rank 37	.339 0 ceiling best median
	Sample stimuli cannot be displayed publicly for this benchmark.

How to use

from brainscore_vision import load_model
model = load_model("vit_large_patch14_clip_224:openai_ft_in1k")
model.start_task(...)
model.start_recording(...)
model.look_at(...)

Model API

Code examples

Curated metadata

At-a-glance

Architecture: Vision Transformer (ViT-L/14), CLIP image encoder
Parameters: ≈304.2M
Input resolution: 224 × 224 pixels
Recurrent: No
Supervision: Contrastive pretrain + supervised fine-tune
Weights provider: Hugging Face Hub, timm org
Trainable layers Probable: show confidence source Probable — https://brain-score.org: 24 transformer blocks; Full fine-tune assumed
Checkpoint: timm/vit_large_patch14_clip_224.openai_ft_in1k

Training and data

CLIP contrastive pretrain on WIT-400M -> direct supervised fine-tune on ImageNet-1k (no IN-12k stage)

Pretrain WIT-400M (OpenAI)

~400M pairs

Finetune ImageNet-1k

1.28M train/50K val

Objective: Probable: show confidence source Probable — curation workbook: CLIP contrastive loss + classification loss (fine-tune)
Loss: Probable: show confidence source Probable — curation workbook: Contrastive (pretrain) + cross-entropy (fine-tune, inferred)
Learning rate:: Not documented
Batch size:: Not documented
Preprocessing: Probable: show confidence source Probable — curation workbook: 224x224, crop_pct 1.0, bicubic; OpenAI CLIP normalization stats

Creators and licensing

Organization:: OpenAI (base pretrain); Ross Wightman/timm (fine-tune, hosting)
Creators:: OpenAI (base CLIP); Ross Wightman (fine-tune)
License: Probable: show confidence source Probable — curation workbook: Apache-2.0 (HF tag), same OpenAI MIT-code / research-intent nuance

Dashed Not documented fields are awaiting curation.

Intended use and limits

Recommended applications

ImageNet classification
transfer-learning backbone/embeddings, lower-compute 224px option

Target users

Probable: show confidence source
Probable — curation workbook

CV researchers/engineers needing compact (224px, lower-compute) ViT-L classifier

Known limitations

Same CLIP-inherited weaknesses (fine-grained classification, counting)
slightly lower top-1 than IN12k->IN1k variant (87.86% vs 88.16%)

Biases

Same FairFace-documented racial/gender disparities & WIT-400M demographic skew as OpenAI CLIP model card

Brain Encoding Response Generator (BERG)

Through the BERG you can easily generate neural responses to images of your choice using any Brain-Score vision model.

For more information on how to use BERG, see the documentation and tutorial.

Historical trend

Hover the line for a preview; click the chart to pin that month in the sidebar. While pinned, hover does not change it -- use Release or Esc to clear.

Benchmarks BibTeX

@article{allen_massive_2022,
    title = {A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence},
    volume = {25},
    issn = {1097-6256},
    doi = {10.1038/s41593-021-00962-x},
    journal = {Nature Neuroscience},
    author = {Allen, Emily J. and St-Yves, Ghislain and Wu, Yihan and Breedlove, Jesse L.
              and Prince, Jacob S. and Dowdle, Logan T. and Nau, Matthias and Caron, Brad
              and Pestilli, Franco and Charest, Ian and Hutchinson, J. Benjamin
              and Naselaris, Thomas and Kay, Kendrick},
    year = {2022},
    pages = {116--126},
}
        @Article{Freeman2013,
                author={Freeman, Jeremy
                and Ziemba, Corey M.
                and Heeger, David J.
                and Simoncelli, Eero P.
                and Movshon, J. Anthony},
                title={A functional and perceptual signature of the second visual area in primates},
                journal={Nature Neuroscience},
                year={2013},
                month={Jul},
                day={01},
                volume={16},
                number={7},
                pages={974-981},
                abstract={The authors examined neuronal responses in V1 and V2 to synthetic texture stimuli that replicate higher-order statistical dependencies found in natural images. V2, but not V1, responded differentially to these textures, in both macaque (single neurons) and human (fMRI). Human detection of naturalistic structure in the same images was predicted by V2 responses, suggesting a role for V2 in representing natural image structure.},
                issn={1546-1726},
                doi={10.1038/nn.3402},
                url={https://doi.org/10.1038/nn.3402}
                }
        @article{li2026triplen,
    title = {Triple-N dataset: large-scale fMRI-guided dense recordings of nonhuman
             primate neural responses to natural scenes},
    author = {Li, Yipeng and Liu, Xieyi and Li, Wanru and Yang, Jia and Gong, Baoqi
              and Jin, Wei and Gong, Zhengxin and Wang, Kesheng and Luo, Jingqiu
              and Zhao, Zishuo and Bao, Pinglei},
    journal = {Nature Neuroscience},
    year = {2026},
    doi = {10.1038/s41593-026-02322-z},
    url = {https://doi.org/10.1038/s41593-026-02322-z},
}
        @article {Marques2021.03.01.433495,
	author = {Marques, Tiago and Schrimpf, Martin and DiCarlo, James J.},
	title = {Multi-scale hierarchical neural network models that bridge from single neurons in the primate primary visual cortex to object recognition behavior},
	elocation-id = {2021.03.01.433495},
	year = {2021},
	doi = {10.1101/2021.03.01.433495},
	publisher = {Cold Spring Harbor Laboratory},
	abstract = {Primate visual object recognition relies on the representations in cortical areas at the top of the ventral stream that are computed by a complex, hierarchical network of neural populations. While recent work has created reasonably accurate image-computable hierarchical neural network models of those neural stages, those models do not yet bridge between the properties of individual neurons and the overall emergent behavior of the ventral stream. One reason we cannot yet do this is that individual artificial neurons in multi-stage models have not been shown to be functionally similar to individual biological neurons. Here, we took an important first step by building and evaluating hundreds of hierarchical neural network models in how well their artificial single neurons approximate macaque primary visual cortical (V1) neurons. We found that single neurons in certain models are surprisingly similar to their biological counterparts and that the distributions of single neuron properties, such as those related to orientation and spatial frequency tuning, approximately match those in macaque V1. Critically, we observed that hierarchical models with V1 stages that better match macaque V1 at the single neuron level are also more aligned with human object recognition behavior. Finally, we show that an optimized classical neuroscientific model of V1 is more functionally similar to primate V1 than all of the tested multi-stage models, suggesting room for further model improvements with tangible payoffs in closer alignment to human behavior. These results provide the first multi-stage, multi-scale models that allow our field to ask precisely how the specific properties of individual V1 neurons relate to recognition behavior.HighlightsImage-computable hierarchical neural network models can be naturally extended to create hierarchical {\textquotedblleft}brain models{\textquotedblright} that allow direct comparison with biological neural networks at multiple scales {\textendash} from single neurons, to population of neurons, to behavior.Single neurons in some of these hierarchical brain models are functionally similar to single neurons in macaque primate visual cortex (V1)Some hierarchical brain models have processing stages in which the entire distribution of artificial neuron properties closely matches the biological distributions of those same properties in macaque V1Hierarchical brain models whose V1 processing stages better match the macaque V1 stage also tend to be more aligned with human object recognition behavior at their output stageCompeting Interest StatementThe authors have declared no competing interest.},
	URL = {https://www.biorxiv.org/content/early/2021/08/13/2021.03.01.433495},
	eprint = {https://www.biorxiv.org/content/early/2021/08/13/2021.03.01.433495.full.pdf},
	journal = {bioRxiv}
}
        @article{Cavanaugh2002,
            author = {Cavanaugh, James R. and Bair, Wyeth and Movshon, J. A.},
            doi = {10.1152/jn.00692.2001},
            isbn = {0022-3077 (Print) 0022-3077 (Linking)},
            issn = {0022-3077},
            journal = {Journal of Neurophysiology},
            mendeley-groups = {Benchmark effects/Done,Benchmark effects/*Surround Suppression},
            number = {5},
            pages = {2530--2546},
            pmid = {12424292},
            title = {{Nature and Interaction of Signals From the Receptive Field Center and Surround in Macaque V1 Neurons}},
            url = {http://www.physiology.org/doi/10.1152/jn.00692.2001},
            volume = {88},
            year = {2002}
            }
        @article{Freeman2013,
            author = {Freeman, Jeremy and Ziemba, Corey M. and Heeger, David J. and Simoncelli, E. P. and Movshon, J. A.},
            doi = {10.1038/nn.3402},
            issn = {10976256},
            journal = {Nature Neuroscience},
            number = {7},
            pages = {974--981},
            pmid = {23685719},
            publisher = {Nature Publishing Group},
            title = {{A functional and perceptual signature of the second visual area in primates}},
            url = {http://dx.doi.org/10.1038/nn.3402},
            volume = {16},
            year = {2013}
            }
        @article{Schiller1976,
            author = {Schiller, P. H. and Finlay, B. L. and Volman, S. F.},
            doi = {10.1152/jn.1976.39.6.1352},
            issn = {0022-3077},
            journal = {Journal of neurophysiology},
            number = {6},
            pages = {1334--1351},
            pmid = {825624},
            title = {{Quantitative studies of single-cell properties in monkey striate cortex. III. Spatial Frequency}},
            url = {http://www.ncbi.nlm.nih.gov/pubmed/825624},
            volume = {39},
            year = {1976}
            }
        @article{papale_extensive_2025,
	title = {An extensive dataset of spiking activity to reveal the syntax of the ventral stream},
	volume = {113},
	issn = {08966273},
	url = {https://linkinghub.elsevier.com/retrieve/pii/S089662732400881X},
	doi = {10.1016/j.neuron.2024.12.003},
	journal = {Neuron},
	author = {Papale, Paolo and Wang, Feng and Self, Matthew W. and Roelfsema, Pieter R.},
	year = {2025},
}
        @inproceedings{santurkar2019computer,
    title={Computer Vision with a Single (Robust) Classifier},
    author={Shibani Santurkar and Dimitris Tsipras and Brandon Tran and Andrew Ilyas and Logan Engstrom and Aleksander Madry},
    booktitle={ArXiv preprint arXiv:1906.09453},
    year={2019}
}
        @inproceedings{zerbe_laion-fmri_2026,
      title = {{LAION}-{fMRI}: A densely sampled 7T-fMRI dataset providing broad coverage of natural image diversity},
      author = {Zerbe, Josefine and Roth, Johannes and Mell, Maggie Mae and Herholz, Peer and Knapen, Tomas and Hebart, Martin N.},
      year = {2026},
      booktitle = {Vision Sciences Society Annual Meeting},
  }
        @article{cowley2026compact,
    title={Compact deep neural network models of the visual cortex},
    author={Cowley, Benjamin R and Stan, Patricia L and Pillow, Jonathan W and Smith, Matthew A},
    journal={Nature},
    volume={652},
    number={8111},
    pages={947--954},
    year={2026},
    publisher={Nature Publishing Group},
    doi={10.1038/s41586-026-10150-1},
    url={https://doi.org/10.1038/s41586-026-10150-1}}
        @article{muzellec_reverse_2026,
      title = {Reverse predictivity for bidirectional comparison of neural networks and biological brains},
      volume = {8},
      issn = {2522-5839},
      url = {https://doi.org/10.1038/s42256-026-01204-0},
      doi = {10.1038/s42256-026-01204-0},
      number = {3},
      journal = {Nature Machine Intelligence},
      author = {Muzellec, Sabine and Kar, Kohitij},
      month = mar,
      year = {2026},
      pages = {474--488},
}
        @article {Majaj13402,
            author = {Majaj, Najib J. and Hong, Ha and Solomon, Ethan A. and DiCarlo, James J.},
            title = {Simple Learned Weighted Sums of Inferior Temporal Neuronal Firing Rates Accurately Predict Human Core Object Recognition Performance},
            volume = {35},
            number = {39},
            pages = {13402--13418},
            year = {2015},
            doi = {10.1523/JNEUROSCI.5181-14.2015},
            publisher = {Society for Neuroscience},
            abstract = {To go beyond qualitative models of the biological substrate of object recognition, we ask: can a single ventral stream neuronal linking hypothesis quantitatively account for core object recognition performance over a broad range of tasks? We measured human performance in 64 object recognition tests using thousands of challenging images that explore shape similarity and identity preserving object variation. We then used multielectrode arrays to measure neuronal population responses to those same images in visual areas V4 and inferior temporal (IT) cortex of monkeys and simulated V1 population responses. We tested leading candidate linking hypotheses and control hypotheses, each postulating how ventral stream neuronal responses underlie object recognition behavior. Specifically, for each hypothesis, we computed the predicted performance on the 64 tests and compared it with the measured pattern of human performance. All tested hypotheses based on low- and mid-level visually evoked activity (pixels, V1, and V4) were very poor predictors of the human behavioral pattern. However, simple learned weighted sums of distributed average IT firing rates exactly predicted the behavioral pattern. More elaborate linking hypotheses relying on IT trial-by-trial correlational structure, finer IT temporal codes, or ones that strictly respect the known spatial substructures of IT ({	extquotedblleft}face patches{	extquotedblright}) did not improve predictive power. Although these results do not reject those more elaborate hypotheses, they suggest a simple, sufficient quantitative model: each object recognition task is learned from the spatially distributed mean firing rates (100 ms) of \~{}60,000 IT neurons and is executed as a simple weighted sum of those firing rates.SIGNIFICANCE STATEMENT We sought to go beyond qualitative models of visual object recognition and determine whether a single neuronal linking hypothesis can quantitatively account for core object recognition behavior. To achieve this, we designed a database of images for evaluating object recognition performance. We used multielectrode arrays to characterize hundreds of neurons in the visual ventral stream of nonhuman primates and measured the object recognition performance of \&gt;100 human observers. Remarkably, we found that simple learned weighted sums of firing rates of neurons in monkey inferior temporal (IT) cortex accurately predicted human performance. Although previous work led us to expect that IT would outperform V4, we were surprised by the quantitative precision with which simple IT-based linking hypotheses accounted for human behavior.},
            issn = {0270-6474},
            URL = {https://www.jneurosci.org/content/35/39/13402},
            eprint = {https://www.jneurosci.org/content/35/39/13402.full.pdf},
            journal = {Journal of Neuroscience}}
        @misc{Sanghavi_DiCarlo_2021,
  title={Sanghavi2020},
  url={osf.io/chwdk},
  DOI={10.17605/OSF.IO/CHWDK},
  publisher={OSF},
  author={Sanghavi, Sachi and DiCarlo, James J},
  year={2021},
  month={Nov}
}
        @misc{Sanghavi_Jozwik_DiCarlo_2021,
  title={SanghaviJozwik2020},
  url={osf.io/fhy36},
  DOI={10.17605/OSF.IO/FHY36},
  publisher={OSF},
  author={Sanghavi, Sachi and Jozwik, Kamila M and DiCarlo, James J},
  year={2021},
  month={Nov}
}
        @misc{Sanghavi_Murty_DiCarlo_2021,
  title={SanghaviMurty2020},
  url={osf.io/fchme},
  DOI={10.17605/OSF.IO/FCHME},
  publisher={OSF},
  author={Sanghavi, Sachi and Murty, N A R and DiCarlo, James J},
  year={2021},
  month={Nov}
}
        @article{gifford_large_2022,
	title = {A large and rich {EEG} dataset for modeling human visual object recognition},
	volume = {264},
	issn = {10538119},
	url = {https://linkinghub.elsevier.com/retrieve/pii/S1053811922008758},
	doi = {10.1016/j.neuroimage.2022.119754},
	journal = {NeuroImage},
	author = {Gifford, Alessandro T. and Dwivedi, Kshitij and Roig, Gemma and Cichy, Radoslaw M.},
	year = {2022},
}
        @Article{Kar2019,
                                                    author={Kar, Kohitij
                                                    and Kubilius, Jonas
                                                    and Schmidt, Kailyn
                                                    and Issa, Elias B.
                                                    and DiCarlo, James J.},
                                                    title={Evidence that recurrent circuits are critical to the ventral stream's execution of core object recognition behavior},
                                                    journal={Nature Neuroscience},
                                                    year={2019},
                                                    month={Jun},
                                                    day={01},
                                                    volume={22},
                                                    number={6},
                                                    pages={974-983},
                                                    abstract={Non-recurrent deep convolutional neural networks (CNNs) are currently the best at modeling core object recognition, a behavior that is supported by the densely recurrent primate ventral stream, culminating in the inferior temporal (IT) cortex. If recurrence is critical to this behavior, then primates should outperform feedforward-only deep CNNs for images that require additional recurrent processing beyond the feedforward IT response. Here we first used behavioral methods to discover hundreds of these `challenge' images. Second, using large-scale electrophysiology, we observed that behaviorally sufficient object identity solutions emerged {	extasciitilde}30{	hinspace}ms later in the IT cortex for challenge images compared with primate performance-matched `control' images. Third, these behaviorally critical late-phase IT response patterns were poorly predicted by feedforward deep CNN activations. Notably, very-deep CNNs and shallower recurrent CNNs better predicted these late IT responses, suggesting that there is a functional equivalence between additional nonlinear transformations and recurrence. Beyond arguing that recurrent circuits are critical for rapid object identification, our results provide strong constraints for future recurrent model development.},
                                                    issn={1546-1726},
                                                    doi={10.1038/s41593-019-0392-5},
                                                    url={https://doi.org/10.1038/s41593-019-0392-5}
                                                    }
        @article {Rajalingham240614,
                author = {Rajalingham, Rishi and Issa, Elias B. and Bashivan, Pouya and Kar, Kohitij and Schmidt, Kailyn and DiCarlo, James J.},
                title = {Large-scale, high-resolution comparison of the core visual object recognition behavior of humans, monkeys, and state-of-the-art deep artificial neural networks},
                elocation-id = {240614},
                year = {2018},
                doi = {10.1101/240614},
                publisher = {Cold Spring Harbor Laboratory},
                abstract = {Primates{	extemdash}including humans{	extemdash}can typically recognize objects in visual images at a glance even in the face of naturally occurring identity-preserving image transformations (e.g. changes in viewpoint). A primary neuroscience goal is to uncover neuron-level mechanistic models that quantitatively explain this behavior by predicting primate performance for each and every image. Here, we applied this stringent behavioral prediction test to the leading mechanistic models of primate vision (specifically, deep, convolutional, artificial neural networks; ANNs) by directly comparing their behavioral signatures against those of humans and rhesus macaque monkeys. Using high-throughput data collection systems for human and monkey psychophysics, we collected over one million behavioral trials for 2400 images over 276 binary object discrimination tasks. Consistent with previous work, we observed that state-of-the-art deep, feed-forward convolutional ANNs trained for visual categorization (termed DCNNIC models) accurately predicted primate patterns of object-level confusion. However, when we examined behavioral performance for individual images within each object discrimination task, we found that all tested DCNNIC models were significantly non-predictive of primate performance, and that this prediction failure was not accounted for by simple image attributes, nor rescued by simple model modifications. These results show that current DCNNIC models cannot account for the image-level behavioral patterns of primates, and that new ANN models are needed to more precisely capture the neural mechanisms underlying primate object vision. To this end, large-scale, high-resolution primate behavioral benchmarks{	extemdash}such as those obtained here{	extemdash}could serve as direct guides for discovering such models.SIGNIFICANCE STATEMENT Recently, specific feed-forward deep convolutional artificial neural networks (ANNs) models have dramatically advanced our quantitative understanding of the neural mechanisms underlying primate core object recognition. In this work, we tested the limits of those ANNs by systematically comparing the behavioral responses of these models with the behavioral responses of humans and monkeys, at the resolution of individual images. Using these high-resolution metrics, we found that all tested ANN models significantly diverged from primate behavior. Going forward, these high-resolution, large-scale primate behavioral benchmarks could serve as direct guides for discovering better ANN models of the primate visual system.},
                URL = {https://www.biorxiv.org/content/early/2018/02/12/240614},
                eprint = {https://www.biorxiv.org/content/early/2018/02/12/240614.full.pdf},
                journal = {bioRxiv}
            }
        @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}
        }
        @article{BAKER2022104913,
                title = {Deep learning models fail to capture the configural nature of human shape perception},
                journal = {iScience},
                volume = {25},
                number = {9},
                pages = {104913},
                year = {2022},
                issn = {2589-0042},
                doi = {https://doi.org/10.1016/j.isci.2022.104913},
                url = {https://www.sciencedirect.com/science/article/pii/S2589004222011853},
                author = {Nicholas Baker and James H. Elder},
                keywords = {Biological sciences, Neuroscience, Sensory neuroscience},
                abstract = {Summary
                A hallmark of human object perception is sensitivity to the holistic configuration of the local shape features of an object. Deep convolutional neural networks (DCNNs) are currently the dominant models for object recognition processing in the visual cortex, but do they capture this configural sensitivity? To answer this question, we employed a dataset of animal silhouettes and created a variant of this dataset that disrupts the configuration of each object while preserving local features. While human performance was impacted by this manipulation, DCNN performance was not, indicating insensitivity to object configuration. Modifications to training and architecture to make networks more brain-like did not lead to configural processing, and none of the networks were able to accurately predict trial-by-trial human object judgements. We speculate that to match human configural sensitivity, networks must be trained to solve a broader range of object tasks beyond category recognition.}
        }
        @misc{ferguson_ngo_lee_dicarlo_schrimpf_2024,
         title={How Well is Visual Search Asymmetry predicted by a Binary-Choice, Rapid, Accuracy-based Visual-search, Oddball-detection (BRAVO) task?},
         url={osf.io/5ba3n},
         DOI={10.17605/OSF.IO/5BA3N},
         publisher={OSF},
         author={Ferguson, Michael E, Jr and Ngo, Jerry and Lee, Michael and DiCarlo, James and Schrimpf, Martin},
         year={2024},
         month={Jun}
}
        @article {Maniquet2024.04.02.587669,
	author = {Maniquet, Tim and de Beeck, Hans Op and Costantino, Andrea Ivan},
	title = {Recurrent issues with deep neural network models of visual recognition},
	elocation-id = {2024.04.02.587669},
	year = {2024},
	doi = {10.1101/2024.04.02.587669},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2024/04/10/2024.04.02.587669},
	eprint = {https://www.biorxiv.org/content/early/2024/04/10/2024.04.02.587669.full.pdf},
	journal = {bioRxiv}
}
        @INPROCEEDINGS{5206848,  
                                                author={J. {Deng} and W. {Dong} and R. {Socher} and L. {Li} and  {Kai Li} and  {Li Fei-Fei}},  
                                                booktitle={2009 IEEE Conference on Computer Vision and Pattern Recognition},   
                                                title={ImageNet: A large-scale hierarchical image database},   
                                                year={2009},  
                                                volume={},  
                                                number={},  
                                                pages={248-255},
                                            }
        @ARTICLE{Hendrycks2019-di,
   title         = "Benchmarking Neural Network Robustness to Common Corruptions
                    and Perturbations",
   author        = "Hendrycks, Dan and Dietterich, Thomas",
   abstract      = "In this paper we establish rigorous benchmarks for image
                    classifier robustness. Our first benchmark, ImageNet-C,
                    standardizes and expands the corruption robustness topic,
                    while showing which classifiers are preferable in
                    safety-critical applications. Then we propose a new dataset
                    called ImageNet-P which enables researchers to benchmark a
                    classifier's robustness to common perturbations. Unlike
                    recent robustness research, this benchmark evaluates
                    performance on common corruptions and perturbations not
                    worst-case adversarial perturbations. We find that there are
                    negligible changes in relative corruption robustness from
                    AlexNet classifiers to ResNet classifiers. Afterward we
                    discover ways to enhance corruption and perturbation
                    robustness. We even find that a bypassed adversarial defense
                    provides substantial common perturbation robustness.
                    Together our benchmarks may aid future work toward networks
                    that robustly generalize.",
   month         =  mar,
   year          =  2019,
   archivePrefix = "arXiv",
   primaryClass  = "cs.LG",
   eprint        = "1903.12261",
   url           = "https://arxiv.org/abs/1903.12261"
}
        @article{hermann2020origins,
              title={The origins and prevalence of texture bias in convolutional neural networks},
              author={Hermann, Katherine and Chen, Ting and Kornblith, Simon},
              journal={Advances in Neural Information Processing Systems},
              volume={33},
              pages={19000--19015},
              year={2020},
              url={https://proceedings.neurips.cc/paper/2020/hash/db5f9f42a7157abe65bb145000b5871a-Abstract.html}
        }

Evaluation and I/O

Test datasets: ImageNet-1k val - top-1 87.86%, top-5 98.43% (results-imagenet.csv)
Validation datasets Probable: show confidence source Probable — curation workbook: ImageNet-1k val
Input format: RGB (B,3,224,224)
Output format: 1000-way logits or (B,257,1024) unpooled tokens / pooled embedding
Tokenizer: Not documented
Model version Probable: show confidence source Probable — curation workbook: Original OpenAI CLIP ViT-L/14 (224px), released January 2022

metadata.yaml · Schema v2.0

This annotated template shows the structure of a model’s metadata.yaml. Each section groups related information; comments explain the expected types and allowed values.

Required: schema_version, domain, and models keyed by model identifier. Other fields may be omitted or set to null when undocumented. Use [] for empty lists.

The values below are illustrative, not this model’s metadata. Metadata changes are proposed in repository pull requests and published after merge.

# metadata.yaml — proposed schema v2.0
# Example values below are illustrative, not metadata for a real model.
# Required: schema_version, domain, and models keyed by exact model identifier.
# Other fields may be omitted or set to null when undocumented.
# Use [] for an empty list. Do not use "unknown" as a numeric/boolean value.
# Keep identifiers exact, including checkpoint suffixes.

schema_version: "2.0"  # String; quote the version.

domain: vision  # Must match the repository's registered domain.
models:
  example-model:
    sources:
      example-source:
        kind: other  # paper | huggingface | other | unreviewed
        url: https://example.org/model-card
        citation: Illustrative source, not evidence for a real model.
    model:
      display_name: Example model
      version: null  # String, e.g. "1.0".
      architecture:
        family: convolutional_neural_network  # String describing the architecture family.
        description: A convolutional model trained for image classification.
      parameter_count: 25000000  # Nonnegative integer; count, not "25M".
      parameter_count_exact: false  # Boolean; false means the count is approximate.
      trainable_layers: "All layers"  # Text; can describe partial fine-tuning.
      recurrent: false  # Boolean.
      input_resolution:
        channels: 3  # Nonnegative integer.
        height: 224  # Nonnegative integer, pixels.
        width: 224  # Nonnegative integer, pixels.
      visual_degrees: null  # Nonnegative number; vision models only.
      visual_degrees_description: null  # Text explaining the field of view.
      supervision:
        type: supervised  # String, e.g. supervised, self_supervised, weakly_supervised.
        description: Trained using class labels.

    training:
      process: null  # Text describing the training procedure.
      objective: null  # Text describing what the model learns.
      loss: null  # Text, e.g. "Cross-entropy".
      learning_rate: null  # Text; supports schedules, e.g. "0.1 with cosine decay".
      batch_size: null  # Text; supports different values across training stages.
      preprocessing: null  # Text describing normalization and augmentation.

    data:
      summary: null  # Text summarizing the data used.
      training_datasets:
        - identifier: example-training-set  # Optional dataset identifier.
          name: Example training set  # Required for each dataset entry.
          role: training  # training | pretraining | fine_tuning
          description: null  # Text describing this dataset's use.

    eval:
      test_datasets: []  # Entries: {identifier: optional, name: required, description: optional}.
      validation_datasets: []  # Same entry structure; role is implied by the list.

    io:
      modality: image  # String, e.g. image or text.
      interface: null  # Text describing the model's interface.
      input_format: null  # Text describing accepted inputs and their shape.
      output_format: null  # Text describing returned values and their shape.
      tokenizer: null  # Text; language models, where applicable.

    provenance:
      weights_provider: null  # Organization or repository providing the weights.
      checkpoint: null  # Exact checkpoint identifier.
      source_url: null  # Link to the original code or model card.
      curation_confidence: null  # Text summarizing confidence in the curated record.

    legal:
      license: null  # License identifier or full terms; distinguish code and weights.

    people:
      creators: []  # List of names, e.g. ["Example Researcher"].
      organizations: []  # List of organization names.

    use:
      applications: []  # List of intended applications (strings).
      users: []  # List of intended user groups (strings).
      limitations: []  # List of known limitations (strings).
      biases: []  # List of known biases (strings).

    lineage:
      base_models:
        - identifier: example-base  # Optional; exact identifier in the same domain.
          name: Example base model  # Required for each base-model entry.
          relationship: fine_tuned_from  # variant_of | fine_tuned_from | derived_from

    # Optional verification records. Omit this section if no verification is recorded.
    # Paths refer to fields in this YAML document, not database columns.
    assertions:
      - path: /model/architecture
        status: probable  # verified | probable | uncertain | undocumented
        sources: [example-source]  # References the structured sources above.

Region	Layer
V1	`blocks.6`
V2	`blocks.7.attn`
V4	`blocks.6.attn`
IT	`blocks.10.attn`

CLIP ViT-L/14
vit_large_patch14_clip_224:openai_ft_in1k

Related variants