The research literature on medical images, driven by the fast growth of deep learning, has proved their robust-ness while dealing with medical datasets. But the available methods meet difficulties in reaching an efficient speed and memory cost. In this work, we present our method used in the MICCAI 2021 FLARE Challenge, which aims to solve the requirements of generalization on unseen data and efficiency which is evaluated by computation time and memory cost. Our work focuses mainly on three points, i.e, segmentation model, data augmentation and post-processing. (1) Efficient 2D segmentation methods can balance accuracy and efficiency. (2) Using augmented data can help the model generalize better. (3) Floating-point number format conversion and resizing images significantly improve the result. With these techniques, we yielded sixth place in the competition. Codes are available at https://github.com/quoccuonglqd/mmsegmentation.