RepBlend reduces modality collapse in multimodal dataset distillation via representation blending and symmetric projection trajectory matching, and beats prior methods on Flickr-30K and MS-COCO.
High-performance large-scale image recognition without normalization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Beyond Modality Collapse: Representations Blending for Multimodal Dataset Distillation
RepBlend reduces modality collapse in multimodal dataset distillation via representation blending and symmetric projection trajectory matching, and beats prior methods on Flickr-30K and MS-COCO.