Weicong Liang

This paper examines pre-training methods to improve DETR model performance in object detection.

A new text encoder tackles challenges in visual text rendering.

New methods enhance text rendering quality in multiple languages.