Inverse design of generic metasurfaces for multifunctional wavefront shaping based on deep neural networks

•Inverse design of the metasurface with large geometry complexity is done by a convolution neural network given multiple target optical responses.•The metasurface unit cell is composed of 10 × 10 pixels for possible generation of diversified patterns to meet various optical requirements.•Metasurfaces for holographic imaging are validated at different frequencies and orthogonal polarization for multifunctional manipulation.