CenterNet is a keypoint-based object detection method that improves upon CornerNet by using a triplet of keypoints (one center and two corners) to detect objects. This approach enhances both precision and recall. The method introduces two modules: center pooling, which enriches center keypoint information, and cascade corner pooling, which enhances corner information by considering both boundary and internal directions. CenterNet achieves an AP of 47.0% on the MS-COCO dataset, outperforming existing one-stage detectors by at least 4.9%. It also demonstrates performance comparable to top two-stage detectors, with an inference speed of 270ms for a 52-layer backbone and 340ms for a 104-layer backbone. The method effectively reduces incorrect bounding boxes by leveraging central region information, leading to improved detection accuracy. CenterNet is efficient and effective, with results showing significant improvements in detecting small and medium objects. The approach is validated through experiments on the MS-COCO dataset, demonstrating its effectiveness in reducing false positives and improving detection performance.CenterNet is a keypoint-based object detection method that improves upon CornerNet by using a triplet of keypoints (one center and two corners) to detect objects. This approach enhances both precision and recall. The method introduces two modules: center pooling, which enriches center keypoint information, and cascade corner pooling, which enhances corner information by considering both boundary and internal directions. CenterNet achieves an AP of 47.0% on the MS-COCO dataset, outperforming existing one-stage detectors by at least 4.9%. It also demonstrates performance comparable to top two-stage detectors, with an inference speed of 270ms for a 52-layer backbone and 340ms for a 104-layer backbone. The method effectively reduces incorrect bounding boxes by leveraging central region information, leading to improved detection accuracy. CenterNet is efficient and effective, with results showing significant improvements in detecting small and medium objects. The approach is validated through experiments on the MS-COCO dataset, demonstrating its effectiveness in reducing false positives and improving detection performance.