LeVLJEPA: End-to-End Vision-Language Pretraining Without Negatives

Benchmark Model Rank Results
zero-shot-transfer-image-classification-on-fgvc-aircraftLeVLJEPA ViT-B/16 (zero-shot)#4Accuracy: 7.65
linear-probe-image-classification-on-fgvc-aircraftLeVLJEPA ViT-B/16 (linear probe)#5Accuracy: 46.38
linear-probe-semantic-segmentation-on-ade20kLeVLJEPA ViT-B/16 (linear segmentation head)#13mIoU: 23.15
image-classification-on-imagenetLeVLJEPA ViT-B/16 (linear probe)#1089Top 1 Accuracy: 65.42%