CustomNet: Zero-shot Object Customization with Variable-Viewpoints in Text-to-Image Diffusion Models