1 paper
Xingfu Yi, Xiaoxue Yu
Recent real-world image super-resolution (SR) methods often adapt text-to-image (T2I) backbones with ControlNet-style branches or spatial conditioning tokens, which increases memor…