A domain-specific pretrained 3D encoder plus cross-batch embedding accumulation makes CLIP-style alignment of brain MRI and tabular data feasible with 62 scans.
Further, we used initial temperature τ = 1.351 in the CLIP-loss and an embedding space with 512 dimensions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Revisiting CLIP: Efficient Alignment of 3D MRI and Tabular Data using Domain-Specific Foundation Models
A domain-specific pretrained 3D encoder plus cross-batch embedding accumulation makes CLIP-style alignment of brain MRI and tabular data feasible with 62 scans.