Explore alternative token-fusion strategies

Explore token-fusion strategies other than channel-wise concatenation for merging image tokens with pretrained representation tokens in DINO-Forcing.

Background

The implementation concatenates image and DINO representation tokens along the channel dimension and then projects the result to the transformer's hidden dimension. The appendix notes that other fusion operations are possible, including approaches discussed in concurrent work, but the authors use channel concatenation for comparison and explicitly defer systematic exploration of alternatives.

References

Here, we stick to the channel concatenation strategy just for proper comparison and leave the exploration of other types of fusion strategies for future work.

— Dino Forcing Flow Models: Do not denoise what you can predict  (2610.11751 - Ghosh et al., 8 Oct 2026) in Appendix, Section 'Explanation behind our fusing Image and Frozen Representations mechanism'