CARZero: Cross-Attention Alignment for Radiology Zero-Shot Classification

We propose CARZero, a novel framework for radiology zero-shot classification that replaces conventional cosine similarity with a learnable cross-attention alignment strategy. CARZero models the complex semantic relationships between medical image and text modalities, achieving state-of-the-art performance on chest X-ray zero-shot diagnosis and object detection benchmarks.

Venue: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2024) — CCF A, Top Venue in Computer Vision
Role: First Author

Recommended citation: Haoran Lai, et al. "CARZero: Cross-Attention Alignment for Radiology Zero-Shot Classification." CVPR 2024.
Download Paper