EchoCare-CLIP achieves 0.682 paired alignment on a 16K ultrasound image-text corpus but downstream zero-shot classification peaks at 0.709 on BUSI only with partial fine-tuning, while full fine-tuning overfits.
Swin transformer V2: Scaling up capacity and resolution
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Ultrasound Vision-Language Alignment via Contrastive Learning
EchoCare-CLIP achieves 0.682 paired alignment on a 16K ultrasound image-text corpus but downstream zero-shot classification peaks at 0.709 on BUSI only with partial fine-tuning, while full fine-tuning overfits.