34. InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions