We present a novel method to improve the accuracy of the 3D reconstruction of\nclothed human shape from a single image. Recent work has introduced volumetric,\nimplicit and model-based shape learning frameworks for reconstruction of\nobjects and people from one or more images. However, the accuracy and\ncompleteness for reconstruction of clothed people is limited due to the large\nvariation in shape resulting from clothing, hair, body size, pose and camera\nviewpoint. This paper introduces two advances to overcome this limitation:\nfirstly a new synthetic dataset of realistic clothed people, 3DVH; and\nsecondly, a novel multiple-view loss function for training of monocular\nvolumetric shape estimation, which is demonstrated to significantly improve\ngeneralisation and reconstruction accuracy. The 3DVH dataset of realistic\nclothed 3D human models rendered with diverse natural backgrounds is\ndemonstrated to allows transfer to reconstruction from real images of people.\nComprehensive comparative performance evaluation on both synthetic and real\nimages of people demonstrates that the proposed method significantly\noutperforms the previous state-of-the-art learning-based single image 3D human\nshape estimation approaches achieving significant improvement of reconstruction\naccuracy, completeness, and quality. An ablation study shows that this is due\nto both the proposed multiple-view training and the new 3DVH dataset. The code\nand the dataset can be found at the project website:\nhttps://akincaliskan3d.github.io/MV3DH/.\n