We present a new approach for oriented object detection, an anchor-free one-stage detector. This approach, named TricubeNet, represents each object as a 2D Tricube kernel and extracts bounding boxes using appearance-based post-processing. Unlike existing anchor-based oriented object detectors, we can save the computational complexity and the number of hyperparameters by eliminating the anchor box in the network design. In addition, by adopting a heatmap-based detection process instead of the box offset regression, we simply and effectively solve the angle discontinuity problem, which is one of the important problems for oriented object detection. To further boost the performance, we propose some effective techniques for the loss balancing, extracting the rotation-invariant feature, and heatmap refinement. To demonstrate the effectiveness of our TricueNet, we experiment on various tasks for the weakly-occluded oriented object detection. The extensive experimental results show that our TricueNet is highly effective and competitive for oriented object detection. The code is available at https://github.com/qjadud1994/TricubeNet.