A post-hoc calibration method for fine-tuned CLIP that scales logits by a weight derived from the L1 distance between original and fine-tuned CLIP logits reduces calibration error on train and unseen classes.
Task residual for tuning vision-language models
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
-
Contrast-Aware Calibration for Fine-Tuned CLIP: Leveraging Image-Text Alignment
A post-hoc calibration method for fine-tuned CLIP that scales logits by a weight derived from the L1 distance between original and fine-tuned CLIP logits reduces calibration error on train and unseen classes.