Han Hu

We propose a training-free projection-based continual merging method that efficiently combines models incrementally.

We present a prompt based multimodal semantic segmentation on the basis of pertained single-modality RGB model