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