Modeling of Few-Shot animation frame generation based on channel attention

Aiming at the problems of low quality and artifacts in character animation generation due to high complexity of motion capture, small sample size, body occlusion, we introduce a model for few-shot generating animation frames based on channel-attention. This paper designs a generator for generating high-quality animation between given start frame and target frame, using two discriminators to discriminate the generated images. A small number of data samples are expanded using data augmentation techniques, and the expanded samples are used as the training set. Key points are extracted from the animation frames hierarchically with key point extraction methods, and then the information is inputted into the channel-based attention generator model. Animation frames are generated according to the input samples, then the size of the generated image is scaled using the up-sampling method. In experiments, carrying out on four different character animations, the datasets come from Adobe Stock. In addition, to verify the model robustness and generalization, we use two types of datasets for testing. The experimental results show that the model can make full use of the animation feature information contained in the given frames to extract the key points feature in a hierarchical manner effectively and reduce the animation artifacts so that improve the degree of realism of the character animation.