Thank you for your amazing work.I searched online and found that the CLIP-ViT-L/14@336px model divides an image into 14*14=196 patches, and the embedding dimension is 768. In your work the shape of features after CLIP visual encoder is (576,1024). How does it come?
Thank you for your amazing work.I searched online and found that the CLIP-ViT-L/14@336px model divides an image into 14*14=196 patches, and the embedding dimension is 768. In your work the shape of features after CLIP visual encoder is (576,1024). How does it come?