CLEAR achieves 15-20% improvement in few-shot classification accuracy on ImageNet-LT and CIFAR-100-LT datasets?

CLEAR achieves 15-20% improvement in few-shot classification accuracy on ImageNet-LT and CIFAR-100-LT datasets

Uses class-wise trust scoring to dynamically weight expert predictions during inference?

Uses class-wise trust scoring to dynamically weight expert predictions during inference

Maintains competitive overall accuracy while significantly boosting performance on rare classes

Research & Papers

arXiv cs.CV August 13, 2026

⚡New CLEAR framework boosts rare-class accuracy by 20% in imbalanced vision datasets...

Deep Dive

Key Points

CLEAR achieves 15-20% improvement in few-shot classification accuracy on ImageNet-LT and CIFAR-100-LT datasets
Uses class-wise trust scoring to dynamically weight expert predictions during inference
Maintains competitive overall accuracy while significantly boosting performance on rare classes

Solves critical imbalance problem in computer vision, enabling more reliable AI for medical imaging and rare object detection.