List your company
ResearchFormerly WizWand

Image Classification on ImageNet-21K

86.6Top-1 AccuracyNomMer-B
-3.245620.079743.40566.7303Nov 25, 2021Aug 25, 2022May 26, 2023Feb 23, 2024Nov 23, 2024Aug 23, 2025May 24, 2026
Updated 5mo ago

Evaluation results

MethodLinks
NomMer-BResolution=384x384, #param. (M)=73, FLOPs (G)=56.22021.1186.6—
Swin-BResolution=384x384, #param. (M)=88, FLOPs (G)=47.12021.1186.4—
ViL-BResolution=384x384, #param. (M)=56, FLOPs (G)=43.72021.1186.2—
NomMer-BResolution=224x224, #param. (M)=73, FLOPs (G)=17.62021.1185.5—
Swin-BResolution=224x224, #param. (M)=88, FLOPs (G)=15.42021.1185.2—
R-101x3Resolution=384x384, #param. (M)=388, FLOPs (G)=204.62021.1184.4—
ViT-B/16Resolution=384x384, #param. (M)=86, FLOPs (G)=55.42021.1184—
Claude Opus 4.5Inference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0561.6416.16
Claude Opus 4.5Inference Strategy=Baseline2026.0545.48—
GPT-4(v4.1)Inference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0545.18.74
Qwen3-VL-PlusInference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0542.231.64
Qwen2.5-VL-7BInference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0537.8737.66
GPT-4(v4.1)Inference Strategy=Baseline2026.0536.36—
Qwen3-VL-4BInference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0534.4133.45
Qwen3-VL-8BInference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0534.223.64
Kimi-K2.5Inference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0532.2616.44
Qwen3-VL-2BInference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0519.2918.81
Kimi-K2.5Inference Strategy=Baseline2026.0515.82—
LLaMA-4Inference Strategy=Divide-and-Conquer Inference (DCI), K=1002026.0514.283.14
LLaMA-4Inference Strategy=Baseline2026.0511.14—
Qwen3-VL-PlusInference Strategy=Baseline2026.0510.56—
Qwen3-VL-8BInference Strategy=Baseline2026.0510.56—
Qwen3-VL-4BInference Strategy=Baseline2026.050.96—
Qwen3-VL-2BInference Strategy=Baseline2026.050.48—
Qwen2.5-VL-7BInference Strategy=Baseline2026.050.21—