In this paper, we propose a two-stage deep convolutional neural network(DCNN) that combines a 2D U-Net-based meniscus localization network with a conditional generative adversarial network-based segmentation network with uncertainty estimation. First, 2D U-Net segments knee MR images into six classes including bones and cartilages using whole MR images with 512x512 resolution to localize the medial and lateral meniscus. Second, adversarial learning using an object-aware map with a generator based on 2D U-Net and discriminator based on 2D DCNN with uncertainty estimation applied to loss function segment the meniscus in localized Region-of-Interests with 64x64 resolution. Dice similarity coefficient between the proposed method and manual segmentation was 86.24% of the medial meniscus and 85.17% for lateral meniscus and showed better results of 6.25%p for medial meniscus and 5.98%p for lateral meniscus compared to the 2D U-Net in the whole area. The proposed method can be used to identify and analyze the shape of the meniscus for allograft transplantation using 3D reconstruction implant model of the patient's un-ruptured meniscus.