List your company

Classification on COCO (subset 300 images, 30 categories)

98AccuracyClean
39.7654.887085.12Mar 29, 2026

Evaluation results

MethodLinks
CleanVisual Encoder=Meta-CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03983.65
CleanVisual Encoder=EVA-CLIP ViT-G/14, Evaluation Protocol=Zero-shot2026.03983.65
CleanVisual Encoder=OpenCLIP ViT-B/16, Evaluation Protocol=Zero-shot2026.03973.65
Natural Light AttackVisual Encoder=Meta-CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03972.14
Natural Light AttackVisual Encoder=EVA-CLIP ViT-G/14, Evaluation Protocol=Zero-shot2026.03972.16
Shadow AttackVisual Encoder=EVA-CLIP ViT-G/14, Evaluation Protocol=Zero-shot2026.03952.03
Natural Light AttackVisual Encoder=OpenCLIP ViT-B/16, Evaluation Protocol=Zero-shot2026.03942.32
CleanVisual Encoder=OpenAI CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03933.65
Natural Light AttackVisual Encoder=OpenAI CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03932.07
Shadow AttackVisual Encoder=OpenCLIP ViT-B/16, Evaluation Protocol=Zero-shot2026.03842.11
ITAVisual Encoder=EVA-CLIP ViT-G/14, Evaluation Protocol=Zero-shot2026.03842.17
Shadow AttackVisual Encoder=Meta-CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03821.91
Shadow AttackVisual Encoder=OpenAI CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03791.78
Ours (random)Visual Encoder=EVA-CLIP ViT-G/14, Evaluation Protocol=Zero-shot2026.03752.89
Ours (random)Visual Encoder=Meta-CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03682.47
Ours (best)Visual Encoder=EVA-CLIP ViT-G/14, Evaluation Protocol=Zero-shot2026.03662.72
ITAVisual Encoder=Meta-CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03642.15
Ours (best)Visual Encoder=Meta-CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03582.39
Ours (random)Visual Encoder=OpenAI CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03532.64
Ours (random)Visual Encoder=OpenCLIP ViT-B/16, Evaluation Protocol=Zero-shot2026.03522.71
ITAVisual Encoder=OpenAI CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03512.15
ITAVisual Encoder=OpenCLIP ViT-B/16, Evaluation Protocol=Zero-shot2026.03462.11
Ours (best)Visual Encoder=OpenCLIP ViT-B/16, Evaluation Protocol=Zero-shot2026.03432.29
Ours (best)Visual Encoder=OpenAI CLIP ViT-L/14, Evaluation Protocol=Zero-shot2026.03422.51