Zero-shot Image Classification on ImageNet Robustness Variants and ObjectNet
75.6Average Class AccuracyViT-L/14 Sparse
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ViT-L/14 SparseSparsity=0.66%, Evaluation=Zero-shot2026.01 | 75.6 | 77.1 | 70.2 | 77.2 | 71.9 | 91 | 66.5 | |
| ViT-L/14 baselineSparsity=100%, Evaluation=Zero-shot2026.01 | 75.1 | 78 | 71.5 | 75.9 | 67.8 | 90.6 | 67.1 | |
| ViT-L/14 Sparse+Sparsity=0.47%, Evaluation=Zero-shot2026.01 | 75.1 | 77 | 69.6 | 76.8 | 70.9 | 90.7 | 66.1 | |
| OpenAI ViT-B/32Sparsity=100%, Evaluation=Zero-shot2026.01 | 57.5 | 68.8 | 60.1 | 53.5 | 29.4 | 77.8 | 56.4 | |
| Prisma ViT-B/32 cls@11Sparsity=2.8%, Evaluation=Zero-shot2026.01 | 56.6 | 67.5 | 59.2 | 52.4 | 28 | 76.8 | 55.7 |