Open-Vocabulary Object Detection Performance in Low-Quality Images

This paper addresses a practical and important problem: evaluating the robustness of open-vocabulary object detection models to low-quality images. The study's significance lies in its focus on real-world image degradation, which is crucial for deploying these models in practical applications. The introduction of a new dataset simulating low-quality images is a valuable contribution, enabling more realistic and comprehensive evaluations. The findings highlight the varying performance of different models under different degradation levels, providing insights for future research and model development.