Pinned post
Today's open source models are GOOD in their domain.
Multi-modal transformers do an excellent job at combining information (DETR, CLIP, etc).
Why not use pretrained models as encoders, and train on their embeddings?
I propose Video Pretrained Transformer. More to come.
Multi-modal transformers do an excellent job at combining information (DETR, CLIP, etc).
Why not use pretrained models as encoders, and train on their embeddings?
I propose Video Pretrained Transformer. More to come.