Sparse pruning method for human body posture estimation model

The invention discloses a sparse pruning method for a human body posture estimation model. According to the method, L1 regularization is used for carrying out sparse training on the human body posture estimation model. Then, the sparse weight of the filter is combined with a scaling factor of a BN (Batch Normalization) layer, so that the importance of the sparse weight of the filter and the scaling factor of the BN layer can be determined; and finally, pruning the sparse filter and the corresponding channel by using a structured pruning method. The method sparsely trains the model to obtain a sparse network structure and prunes unimportant filters by combining sparsity of the convolutional layer and the BN layer. According to the method, the model can be compressed and accelerated while the human body posture estimation precision is guaranteed, and the model can be conveniently deployed on resource-limited mobile terminal equipment. 本发明公开了一种人体姿态估计模型稀疏化剪枝方法，该方法使用L1正则化对人体姿态估计模型进行稀疏训练。然后，将滤波器的稀疏权重与BN(Batch Norma