Since the structure of the flower is too complex, the modeling of the flower faces huge challenges. This paper collects 3D scenes containing flower models, uses 3dsMax to extract flower models, and constructs flower dataset. This paper proposes an encoder-decoder network structure called MVF3D and adopted the trained MVF3D network to predict the missing perspective information, use a single RGB image to generate a depth map of flowers from different perspectives, and finally use the depth maps to reconstruct the flower models. To evaluate the performance of our proposed method, for simple flowers, the average chamfer distance between the reconstructed 3D model and the real model is 0.27, The experimental results have shown that our proposed method can preserve the true structure of the flower.