This study focuses on training and comparing four object detection models—Grounding DINO, RTMDet, Faster R-CNN, and RetinaNet—using the VinDr CXR dataset, containing a diverse set of labeled X-ray images covering various lung diseases. The main objective was to explore the practical aspects of training and evaluating these models while gaining insights into their performance on medical imaging tasks. Metrics such as AUC, AP, Accuracy, Specificity, and Sensitivity were used for evaluation.
