Multi-person Head Segmentation in Low Resolution Crowd Scenes Using Convolutional Encoder-Decoder Framework