DatBench: Discriminative, Faithful, and Efficient VLM Evaluations
A benchmark and evaluation framework for vision-language models built around discriminative tasks, faithful scoring, and practical efficiency.
Publications
Papers and published research
DatBench leads the page as the current benchmark story, followed by earlier papers in multimodal representation learning and medical imaging.
Featured research
DatBench
Current work on making vision-language model evaluation more discriminative, more faithful, and practical enough to run in real research loops.
Research archive
Earlier papers
Earlier work is presented as a quieter reading list, with explicit links out to the paper, code, or publisher page where available.
GraFT: Gradual Fusion Transformer for Multimodal Re-Identification
Transformer-based multimodal re-identification focused on stronger fusion and cleaner representation learning.
