Classification Model Zoo

Benchmarks

Algorithm	Config	Top-1 (%)	Top-5 (%)	gpu memory (MB)	inference time (ms/img)	Download
resnet50(raw)	resnet50(raw)	76.454	93.084	2412	8.59	model
resnet50(tfrecord)	resnet50(tfrecord)	76.266	92.972	2412	8.59	model
resnet101	resnet101	78.152	93.922	2484	16.77	model
resnet152	resnet152	78.544	94.206	2544	24.69	model
resnext50-32x4d	resnext50-32x4d	77.604	93.856	4718	12.88	model
resnext101-32x4d	resnext101-32x4d	78.568	94.344	4792	26.84	model
resnext101-32x8d	resnext101-32x8d	79.468	94.434	9582	27.52	model
resnext152-32x4d	resnext152-32x4d	78.994	94.462	4852	41.08	model
hrnetw18	hrnetw18	76.258	92.976	4701	54.55	model
hrnetw30	hrnetw30	77.66	93.862	4766	54. 30	model
hrnetw32	hrnetw32	77.994	93.976	4780	53.48	model
hrnetw40	hrnetw40	78.142	93.956	4843	54.31	model
hrnetw44	hrnetw44	79.266	94.476	4884	54.83	model
hrnetw48	hrnetw48	79.636	94.802	4916	54.14	model
hrnetw64	hrnetw64	79.884	95.04	5120	54.74	model
vit-base-patch16	vit-base-patch16	76.082	92.026	346	8.03	model
swin-tiny-patch4-window7	swin-tiny-patch4-window7	80.528	94.822	132	12.94	model
deitiii-small-patch16-224	deitiii-small-patch16-224	81.408	95.388	90	7.41 4.90(A100_80G)	model
deitiii-base-patch16-192	deitiii-base-patch16-192	82.982	95.95	337	7.49 5.04(A100_80G)	model
deitiii-large-patch16-192	deitiii-large-patch16-192	83.902	96.296	1170	14.35 9.91(A100_80G)	model
deit_base_patch16_224 (Hydra Attention [8 layers])	deit_base_patch16_224 (Hydra Attention [8 layers])	79.444	94.468	340	6.78 4.47(A100_80G)	model
deit_base_patch16_224 (Hydra Attention [12 layers])	deit_base_patch16_224 (Hydra Attention [12 layers])	76.67	92.872	338	6.65 4.34(A100_80G)	model

(ps: 通过EasyCV训练得到模型结果，推理的输入尺寸默认为224，机器默认为V100 16G，其中gpu memory记录的是gpu peak memory)

Algorithm	Config	Top-1 (%)	Top-5 (%)	gpu memory (MB)	inference time (ms/img)	Download
vit_base_patch16_224	vit_base_patch16_224	78.096	94.324	346	8.03	model
vit_large_patch16_224	vit_large_patch16_224	84.404	97.276	1171	16.30	model
deit_base_patch16_224	deit_base_patch16_224	81.756	95.6	346	7.98	model
deit_base_distilled_patch16_224	deit_base_distilled_patch16_224	83.232	96.476	349	8.07	model
xcit_medium_24_p8_224	xcit_medium_24_p8_224	83.348	96.21	884	31.77	model
xcit_medium_24_p8_224_dist	xcit_medium_24_p8_224_dist	84.876	97.164	884	32.08	model
xcit_large_24_p8_224	xcit_large_24_p8_224	83.986	96.47	1962	37.44	model
xcit_large_24_p8_224_dist	xcit_large_24_p8_224_dist	85.022	97.29	1962	37.44	model
tnt_s_patch16_224	tnt_s_patch16_224	76.934	93.388	100	18.92	model
convit_tiny	convit_tiny	72.954	91.68	31	10.79	model
convit_small	convit_small	81.342	95.784	122	11.23	model
convit_base	convit_base	82.27	95.916	358	11.26	model
cait_xxs24_224	cait_xxs24_224	78.45	94.154	50	22.62	model
cait_xxs36_224	cait_xxs36_224	79.788	94.87	71	33.25	model
cait_s24_224	cait_s24_224	83.302	96.568	190	23.74	model
levit_128	levit_128	78.468	93.874	76	15.33	model
levit_192	levit_192	79.72	94.664	128	15.17	model
levit_256	levit_256	81.432	95.38	222	15.27	model
convnext_tiny	convnext_tiny	81.878	95.836	128	7.17	model
convnext_small	convnext_small	82.836	96.458	213	12.89	model
convnext_base	convnext_base	83.73	96.692	364	13.04	model
convnext_large	convnext_large	84.164	96.844	781	13.78	model
resmlp_12_distilled_224	resmlp_12_distilled_224	77.876	93.532	66	4.90	model
resmlp_24_distilled_224	resmlp_24_distilled_224	80.548	95.204	124	9.07	model
resmlp_36_distilled_224	resmlp_36_distilled_224	80.944	95.416	181	13.56	model
resmlp_big_24_distilled_224	resmlp_big_24_distilled_224	83.45	96.65	534	20.48	model
coat_tiny	coat_tiny	78.112	93.972	127	33.09	model
coat_mini	coat_mini	80.912	95.378	247	33.29	model
convmixer_768_32	convmixer_768_32	80.08	94.992	4995	10.23	model
convmixer_1024_20_ks9_p14	convmixer_1024_20_ks9_p14	81.742	95.578	2407	6.29	model
convmixer_1536_20	convmixer_1536_20	81.432	95.38	547	14.66	model
gmixer_24_224	gmixer_24_224	78.088	93.6	104	11.65	model
gmlp_s16_224	gmlp_s16_224	77.204	93.358	81	11.15	model
mixer_b16_224	mixer_b16_224	72.558	90.068	241	5.37	model
mixer_l16_224	mixer_l16_224	68.34	86.11	804	11.74	model
jx_nest_tiny	jx_nest_tiny	81.278	95.618	90	9.05	model
jx_nest_small	jx_nest_tiny	83.144	96.3	174	16.92	model
jx_nest_base	jx_nest_base	83.474	96.442	300	16.88	model
pit_s_distilled_224	pit_s_distilled_224	83.144	96.3	109	7.00	model
pit_b_distilled_224	pit_b_distilled_224	83.474	96.442	330	7.66	model
twins_svt_small	twins_svt_small	81.598	95.55	657	14.07	model
twins_svt_base	twins_svt_base	82.882	96.234	1447	18.99	model
twins_svt_large	twins_svt_large	83.428	96.506	2567	19.11	model
swin_base_patch4_window7_224	swin_base_patch4_window7_224	84.714	97.444	375	23.47	model
swin_large_patch4_window7_224	swin_large_patch4_window7_224	85.826	97.816	788	23.29	model
dynamic_swin_small_p4_w7_224	dynamic_swin_small_p4_w7_224	82.896	96.234	220	28.55	model
dynamic_swin_tiny_p4_w7_224	dynamic_swin_tiny_p4_w7_224	80.912	95.41	136	14.58	model
shuffletrans_tiny_p4_w7_224	shuffletrans_tiny_p4_w7_224	82.176	96.05	5311	13.90	model
efficientformer_l1	efficientformer_l1	80.102	94.934	1820	7.5	model
efficientformer_l3	efficientformer_l3	82.272	96.028	2436	13.07	model
efficientformer_l7	efficientformer_l7	83.076	96.44	1622	18.96	model
EdgeVit_xxs_b512_224	EdgeVit_xxs_b512_224	75.18	92.188	206	8.67	model
EdgeVit_xs_b256_224	EdgeVit_xs_b256_224	77.624	93.47	551	8.04	model
EdgeVit_s_b128_224	EdgeVit_s_b128_224	80.3	95.302	576	13.49	model

(ps: 通过导入官方模型得到推理结果，需要torch.version >= 1.9.0，推理的输入尺寸默认为224，机器默认为V100 16G，其中gpu memory记录的是gpu peak memory)