Research on human action video generation based on GAN

Aiming at the problems of insufficient action coherence and poor visual realism in existing human action video generation methods, a video generation model based on generative adversarial network (GAN) is proposed. The model significantly improves the accuracy of action representation and visual quality of the generated video by introducing an action encoder, a temporal attention mechanism, and a combination of multiple loss functions. Experiments on the Human3.6M dataset show that the proposed method outperforms the existing methods in terms of FID, PSNR, SSIM, etc., which verifies the effectiveness of the model and provides new ideas and technical support for the subsequent related research.