RGT
RGT enhances image super-resolution by utilizing global spatial information for better reconstruction. It employs recursive-generalization self-attention and cross-attention to improve global context processing. The model minimizes redundancy through scaled attention matrices and integrates features via hybrid adaptive methods. Tests show RGT's effectiveness over recent techniques in various evaluations.
