从两个矩阵中找到形变关系

[lix19937]

 tensor([
        [ 1.6659, -5.5004,  0.2453,  1.6863],
        [ 0.3991, -4.7299,  0.6339,  4.8427],
        [ 0.8844, -1.8214,  0.4939,  1.4171],
        [ 0.9366,  1.7840,  1.2091, -2.9199],
        [-0.0271,  0.1723, -1.4438, -2.5426],
        [-1.0664,  1.6227, -0.0488,  2.7245],
        [-0.6343, -4.7691,  1.3364, 10.4108],
        [-0.3888, -1.4212, -0.3381,  3.3882],
        [-1.4337,  0.0264,  0.6598,  3.0084],
        [ 0.0583, -0.6085,  0.3484,  0.1129],
        [ 0.2647, -0.1973,  0.3074,  0.1219],
        [-2.3718,  1.9050, -0.5367, -0.2446]])

tensor([[[[[ 1.6659,  0.3991,  0.8844,  0.9366],
           [-5.5004, -4.7299, -1.8214,  1.7840]],

          [[-0.0271, -1.0664, -0.6343, -0.3888],
           [ 0.1723,  1.6227, -4.7691, -1.4212]],

          [[-1.4337,  0.0583,  0.2647, -2.3718],
           [ 0.0264, -0.6085, -0.1973,  1.9050]]]],

        [[[[ 0.2453,  0.6339,  0.4939,  1.2091],
           [ 1.6863,  4.8427,  1.4171, -2.9199]],

          [[-1.4438, -0.0488,  1.3364, -0.3381],
           [-2.5426,  2.7245, 10.4108,  3.3882]],

          [[ 0.6598,  0.3484,  0.3074, -0.5367],
           [ 3.0084,  0.1129,  0.1219, -0.2446]]]]])

[lix19937]

permute + permute 融合
可以融合，实际效果可以通过 export onnx then netron