Leaderboard

Below is the leaderboard for our competition, including our baseline FastCLIP.

Name	DataComp Average	ImageNet Zero-Shot Accuracy	Organization	Submission Date	Writeup

Name	DataComp Average	ImageNet Zero-Shot Accuracy	Organization	Submission Date	Writeup

Name	DataComp Average	ImageNet Zero-Shot Accuracy	Organization	Submission Date	Writeup

Name	Model	# Params (M)	Resolution	Dataset	Dataset Size (B)	Batch Size (K)	Samples Seen (B)	ImageNet-1K Top 1	DataComp Average	Weights
EVA-CLIP	ViT-E/14	5045	224	LAION-2B	2	144	9	81.97	69.30	Huggingface
EVA-CLIP	ViT-E/14	4705	224	LAION-2B	2	115	4	81.96	66.66	Huggingface
OpenCLIP	ViT-bigG/14	2540	224	LAION-2B	2	160	36	80.08	66.65	Huggingface
CLIPA	ViT-bigG/14	2518	336	Datacomp-1B	1.4	64	13.4	83.07	68.44	Huggingface
CLIPA	ViT-bigG/14	2517	224	Datacomp-1B	1.4	64	13.3	82.67	68.26	Huggingface
OpenCLIP	ViT-g/14	1367	224	LAION-2B	2	86.7	34.5	78.48	64.26	Huggingface
OpenCLIP	ViT-g/14	1367	224	LAION-2B	2	42	13	76.66	62.87	Huggingface
EVA-CLIP	ViT-g/14	1367	224	Merged-2B	2	114	11	79.32	66.05	Huggingface
OpenCLIP	ConvNeXt-XXL	1201	256	LAION-2B	2	82	34	79.46	65.11	Huggingface
OpenCLIP	ConvNeXt-XXL	1201	256	LAION-2B	2	82	34	79.32	64.96	Huggingface
OpenCLIP	ConvNeXt-XXL	1201	256	LAION-2B	2	82	34	79.07	64.93	Huggingface
OpenCLIP	ViT-H/14	1193	224	LAION-5B	5	90	13	76.97	65.15	Huggingface
EVA-CLIP	ViT-g/14	1136	224	LAION-400M	0.4	41	11	78.52	63.39	Huggingface
DFN	ViT-H/14	987	378	DFN-5B	5	78	44	84.35	70.79	Huggingface
DFN	ViT-H/14	986	224	DFN-5B	5	78	39	83.43	69.62	Huggingface
MetaCLIP	ViT-H/14	986	224	MetaCLIP-2.5B	2.5	32	12.8	80.53	66.72	Meta
OpenCLIP	ViT-H/14	986	224	LAION-2B	2	77.25	34	77.92	64.02	Huggingface
CLIPA	ViT-H/14	969	336	Datacomp-1B	1.4	64	13.4	81.81	66.77	Huggingface
CLIPA	ViT-H/14	969	336	LAION-2B	2	64	13.4	79.10	64.39	Huggingface
CLIPA	ViT-H/14	968	224	Datacomp-1B	1.4	64	13.3	81.50	66.53	Huggingface
ViTamin	ViTamin-XL	925	336	Datacomp-1B	1.4	90	41	82.67	68.12	Huggingface
ViTamin	ViTamin-XL	925	256	Datacomp-1B	1.4	90	40	82.25	67.66	Huggingface
ViTamin	ViTamin-XL	925	384	Datacomp-1B	1.4	90	41	81.39	66.38	Huggingface
SigLIP	ViT-SO400M/14	878	384	WebLI	10	32	45	83.09	69.21	Huggingface
SigLIP	ViT-SO400M/14	877	256	WebLI	10	32	40	82.04	68.07	Huggingface

Name	Model	# Params (M)	Resolution	Dataset	Dataset Size (B)	Batch Size (K)	Samples Seen (B)	ImageNet-1K Top 1	DataComp Average	Weights
SigLIP	ViT-B/16	371	256	WebLI	10	32	40	75.11	60.68	Huggingface
OpenCLIP	ConvNeXt-L	352	320	LAION-2B	2	131	28.5	76.83	63.87	Huggingface
OpenCLIP	ConvNeXt-L	352	320	LAION-2B	2	131	28.5	76.60	63.45	Huggingface
OpenCLIP	ConvNeXt-L	352	256	LAION-2B	2	102	26	75.92	62.95	Huggingface
SigLIP	ViT-B/16	204	512	WebLI	10	32	45	79.13	64.59	Huggingface
SigLIP	ViT-B/16	203	384	WebLI	10	32	45	78.48	63.79	Huggingface
SigLIP	ViT-B/16	203	256	WebLI	10	32	40	76.53	62.26	Huggingface
SigLIP	ViT-B/16	203	224	WebLI	10	32	40	76.04	62.33	Huggingface
MobileCLIP	ViT-B/16	150	224	DatacompDR	1.4	64	39	77.18	65.22	Huggingface
MobileCLIP	ViT-B/16	150	224	DatacompDR	1.4	64	13	76.83	65.03	Huggingface
DFN	ViT-B/16	150	224	DFN-2B	2	78	12.8	76.23	60.75	Huggingface

Name	Model	# Params (M)	Resolution	Dataset	Dataset Size (B)	Batch Size (K)	Samples Seen (B)	ImageNet-1K Top 1	DataComp Average	Weights
ViTamin	ViTamin-L2	688	384	Datacomp-1B	1.4	90	13.3	82.05	67.97	Huggingface
ViTamin	ViTamin-L2	688	336	Datacomp-1B	1.4	90	13.3	81.80	67.72	Huggingface
VITamin	ViTamin-L2	688	256	Datacomp-1B	1.4	90	13.3	81.43	67.20	Huggingface
ViTamin	ViTamin-L2	688	224	Datacomp-1B	1.4	90	12.8	80.89	67.39	Huggingface
SigLIP	ViT-L/16	652	384	WebLI	10	32	45	82.10	66.83	Huggingface
SigLIP	ViT-L/16	652	256	WebLI	10	32	40	80.43	65.56	Huggingface
ViTamin	ViTamin-L	457	384	Datacomp-1B	1.4	90	13	81.83	67.12	Huggingface
VITamin	ViTamin-L	457	336	Datacomp-1B	1.4	90	13	81.60	66.88	Huggingface
ViTamin	ViTamin-L	457	256	Datacomp-1B	1.4	90	13	81.19	66.91	Huggingface
ViTamin	ViTamin-L	457	224	Datacomp-1B	1.4	90	12.8	80.77	66.59	Huggingface
CoCa (OpenCLIP)	ViT-L/14	430	224	LAION-2B	2	90	13	75.64	62.88	Huggingface
EVA-CLIP	ViT-L/14	428	336	Merged-2B	2	61	6	80.37	65.61	Huggingface
OpenAI CLIP	ViT-L/14	428	336	WIT	0.4	32	13.2	76.56	62.76	OpenAI
EVA-CLIP	ViT-L/14	428	224	Merged-2B	2	131	4	79.78	64.76	Huggingface
DFN	ViT-L/14	428	224	DFN-2B	2	88	12.8	81.41	66.65	Huggingface
DataComp	ViT-L/14	428	224	Datacomp-1B	1.4	90	13	79.21	66.28	Huggingface
MetaCLIP	ViT-L/14	428	224	MetaCLIP-2.5B	2.5	32	12.8	79.17	65.92	Meta
DataComp	ViT-L/14	428	224	CommonPool (CLIP Score filtering)	3.8	90	13	76.39	64.80	Huggingface
MetaCLIP	ViT-L/14	428	224	MetaCLIP-400M	0.4	32	12.8	76.17	62.55	Meta
OpenAI CLIP	ViT-L/14	428	224	WIT	0.4	32	12.8	75.56	61.62	OpenAI
DataComp	ViT-L/14	428	224	CommonPool (LAION-2B filtering)	1.3	90	13	75.49	63.56	Huggingface
OpenCLIP	ViT-L/14	428	224	LAION-2B	2	82	32	75.23	61.89	Huggingface
CLIPA-v2	ViT-L/14	415	336	Datacomp-1B	1.4	64	13.4	80.27	65.69	Huggingface
CLIPA-v2	ViT-L/14	414	224	Datacomp-1B	1.4	64	13.3	79.60	65.35	Huggingface