GLOBER: Coherent Non-autoregressive Video Generation via Global Guided Video Decoder

The goal of this work is to advance research on video generation methods. All experiments are conducted without conditional inputs. Figure 1: Genetated long videos with 128 frames on the Sky Time-lapse and UCF-101 datasets (4 frames skipped). A.5 Settings of Hyper Parameters The detailed settings of model hyper parameters are presented in Table 4. Table 4: Hyper-parameters of the video auto-encoder and the quantitative results on video reconstruction. Experimental settings on the UCF-101 dataset are the same for both conditional and unconditional video generation except given video descriptions.