Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
×11{\times}1 projection (C→C/, C\!\to\!C/2), followed by a depth‑wise
1
0
3
By explicitly encoding scene layout priors into the attention mechanism, RPT-SR achieves state-of-the-art infrared image super-resolution, outperforming general-purpose transformers that waste capacity on redundant learning.