The automated analysis of white blood cells (WBCs) and platelets from microscopic images is critical for the diagnosis and monitoring of numerous hematological diseases. While deep learning has significantly advanced this field, progress has been hampered by a notable scarcity of large-scale, publicly available datasets for the task of object detection, in contrast to the relative abundance of datasets for classification. This paper introduces the White Blood Cells and Platelets Detection (WBCPD) dataset, a new, high-quality benchmark created to address this gap. We demonstrate a practical and effective methodology for repurposing an existing, large-scale classification dataset into a resource for high-fidelity object detection through a meticulous manual annotation process. The resulting dataset comprises 9,293 images with fine-grained labels for five distinct WBC subtypes and platelets. We conduct a comprehensive performance evaluation using a suite of state-of-the-art models. Our results show that models trained on WBCPD achieve exceptional performance, with top scores of 0.993 for mAP@0.5 and 0.909 for mAP@0.5:0.9. Comparative analysis shows models trained on WBCPD significantly outperform those on the existing BCCD benchmark, validating its superiority. By releasing WBCPD, we provide a valuable new benchmark to spur further innovation in automated hematology.