689 billion images.
Reusing single-view vision model attention yields scalable correspondence matching
FAST model rewires self-attention layers into cross-attention without pair-centric pretraining, achieving state-of-the-art dense matching.
Independent
Yongjian Zhang · Longguang Wang · Zhuo Song · Zhiheng Fu · Liang Lin · Yulan Guo
Research Digest··2 min read
Zhang et al.
Why this paper
Independent
In one line
A zero-parameter rewiring of self-attention layers in a single-view vision foundation model yields a scalable state-of-the-art correspondence matcher.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§