【论文&模型讲解】ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision