Object detection requires careful handling of two types of features: multi-scale features and multi-task features, referred to as scale-dimension features and task-dimension features in this paper. While numerous studies have focused on enhancing detection performance by fusing features within a single dimension, such as the classic multi-scale fusion pyramid employed by FPN in the scale dimension, limited attention has been given to cross-dimension feature fusion. This paper introduces a comprehensive feature fusion framework, termed Cross-Dimension Feature Fusion (CDFF), which simultaneously fuses features from both scale and task dimensions. Considering the optimal balance between computational efficiency and detection accuracy offered by YOLO models, we implement CDFF based on the latest YOLOv10 detector. CDFF incorporates two key components: (1) an alternate and serial fusion path that spans both the scale and task dimensions, and (2) an efficient multi-path feature fusion module. Within the multi-path fusion module, we introduce adaptive spatial aggregation with cross-attention mechanisms to further improve feature synergy. Extensive experiments on the COCO benchmark demonstrate the effectiveness of the proposed method. Comparisons with state-of-the-art real-time detectors, including YOLOv10, YOLOv11 and RT-DETR, confirm that CDFF achieves the highest AP and AP \(_{75}\) scores and offers a competitive balance between accuracy and efficiency. In addition, fair comparisons demonstrate the generalization of CDFF with various advances in object detection, including YOLO detectors, non-real-time detectors, and transformer backbone. It can serve as an efficient plug-and-play module for approaches like YOLOv11 to enhance their performance, and the code is available at https://github.com/wuguikel/cdff .