MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training
Paper
•
2311.17049
•
Published
MobileCLIP: Mobile-friendly image-text models with SOTA zero-shot capabilities. DataCompDR: Improved datasets for training image-text SOTA models.
Note ^ MobileCLIP checkpoints for the timm library (image-tower only)
Note ^ MobileCLIP checkpoints for the OpenCLIP library
Note ^ MobileCLIP checkpoints, original format
Note ^ DataCompDR datasets