Open-domain visual recognition on INFOSEEK Overall
62.7Top-1 AccuracyWikiCLIP-L
Evaluation Results
| Method | Links | |
|---|---|---|
| WikiCLIP-LFinetune Dataset=OVEN, Text Encoder=LLaMa3.2 3B2026.03 | 62.7 | |
| WikiCLIP-SFinetune Dataset=OVEN, Text Encoder=LLaMa3.2 1B2026.03 | 61.2 | |
| EchosightFinetune Dataset=E-VQA2026.03 | 53.2 | |
| CLIP I2I*Finetune Dataset=-2026.03 | 45.9 | |
| CLIP I2T*Finetune Dataset=-2026.03 | 32 | |
| DPRFinetune Dataset=In-house Data2026.03 | 29.6 |