Python 难以确定2D特征矩阵结构以输入机器学习算法

我正在训练一个情绪识别系统，它通过面部运动来检测情绪。结果，我形成了一个4维矩阵，我试图将其简化为2维

构成4D矩阵的功能：
视频数量（每个视频将被分配情感标签）
每个视频的帧数
每帧面部标志的方向
每帧人脸标记的速度

我尝试训练的重要功能：
左侧是速度（每帧相同面部标志之间的斜边）
右侧是方向（每帧相同面部地标的x和y值的arctan）

我一直使用的4D矩阵，并试图将其简化为2D

>> main.shape  
(60, 17, 68, 2)  
# 60 videos, 17 frames per video, 68 facial landmarks, 2 features (direction and speed)  
>> main  
array([[[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  1.        ,   1.        ],
         [  1.41421356,   0.78539816],
         [  1.41421356,   0.78539816],
         ..., 
         [  3.        ,   1.        ],
         [  3.        ,   1.        ],
         [  3.        ,   1.        ]],

        [[  0.        ,   0.        ],
         [ -1.41421356,   0.78539816],
         [ -1.41421356,   0.78539816],
         ..., 
         [  2.        ,   1.        ],
         [  3.        ,   1.        ],
         [  3.        ,   1.        ]],

        ..., 
        [[  1.        ,   1.        ],
         [  1.41421356,  -0.78539816],
         [  1.41421356,  -0.78539816],
         ..., 
         [ -1.41421356,   0.78539816],
         [  1.        ,   1.        ],
         [ -1.41421356,   0.78539816]],

        [[  2.23606798,  -0.46364761],
         [  2.82842712,  -0.78539816],
         [  2.23606798,  -0.46364761],
         ..., 
         [  1.        ,   0.        ],
         [  0.        ,   0.        ],
         [  1.        ,   1.        ]],

        [[ -1.41421356,  -0.78539816],
         [ -2.23606798,  -0.46364761],
         [ -2.23606798,  -0.46364761],
         ..., 
         [  1.41421356,  -0.78539816],
         [  1.41421356,  -0.78539816],
         [  2.23606798,  -1.10714872]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  2.        ,   1.        ],
         [  2.23606798,  -1.10714872],
         [  1.41421356,  -0.78539816],
         ..., 
         [ -2.        ,  -0.        ],
         [ -1.        ,  -0.        ],
         [ -1.41421356,  -0.78539816]],

        [[  2.        ,   1.        ],
         [ -2.23606798,   1.10714872],
         [ -1.41421356,   0.78539816],
         ..., 
         [  1.        ,   1.        ],
         [ -1.        ,  -0.        ],
         [ -1.        ,  -0.        ]],

        ..., 
        [[ -2.        ,  -0.        ],
         [ -3.        ,  -0.        ],
         [ -4.12310563,  -0.24497866],
         ..., 
         [  0.        ,   0.        ],
         [ -1.        ,  -0.        ],
         [ -2.23606798,   1.10714872]],

        [[ -2.23606798,   1.10714872],
         [ -1.41421356,   0.78539816],
         [ -2.23606798,   1.10714872],
         ..., 
         [ -2.23606798,   0.46364761],
         [ -1.41421356,   0.78539816],
         [ -1.41421356,   0.78539816]],

        [[  2.        ,   1.        ],
         [  1.41421356,   0.78539816],
         [  2.82842712,   0.78539816],
         ..., 
         [  1.        ,   1.        ],
         [  1.        ,   1.        ],
         [ -2.23606798,  -1.10714872]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  1.        ,   1.        ],
         [  0.        ,   0.        ],
         [  1.        ,   1.        ],
         ..., 
         [ -3.        ,  -0.        ],
         [ -2.        ,  -0.        ],
         [  0.        ,   0.        ]],

        [[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  1.41421356,   0.78539816],
         [  1.        ,   0.        ],
         [  0.        ,   0.        ]],

        ..., 
        [[  1.        ,   0.        ],
         [  1.        ,   1.        ],
         [  0.        ,   0.        ],
         ..., 
         [  2.        ,   1.        ],
         [  3.        ,   1.        ],
         [  3.        ,   1.        ]],

        [[ -7.28010989,   1.29249667],
         [ -7.28010989,   1.29249667],
         [ -8.54400375,   1.21202566],
         ..., 
         [-22.02271555,   1.52537305],
         [ 22.09072203,  -1.48013644],
         [ 22.36067977,  -1.39094283]],

        [[  1.        ,   0.        ],
         [  1.41421356,  -0.78539816],
         [  1.        ,   0.        ],
         ..., 
         [ -1.41421356,  -0.78539816],
         [  1.        ,   1.        ],
         [  1.41421356,   0.78539816]]],


       ..., 
       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  5.38516481,   0.38050638],
         [  5.09901951,   0.19739556],
         [  4.47213595,  -0.46364761],
         ..., 
         [ -1.41421356,   0.78539816],
         [ -2.82842712,   0.78539816],
         [ -5.        ,   0.64350111]],

        [[ -6.32455532,   0.32175055],
         [ -6.08276253,  -0.16514868],
         [ -5.65685425,  -0.78539816],
         ..., 
         [  3.60555128,   0.98279372],
         [  5.        ,   0.92729522],
         [  5.65685425,   0.78539816]],

        ..., 
        [[ -3.16227766,  -0.32175055],
         [ -3.60555128,  -0.98279372],
         [  5.        ,   1.        ],
         ..., 
         [ 12.08304597,   1.14416883],
         [ 13.15294644,   1.418147  ],
         [ 14.31782106,   1.35970299]],

        [[  3.60555128,  -0.5880026 ],
         [  4.47213595,  -1.10714872],
         [  6.        ,   1.        ],
         ..., 
         [-20.39607805,   1.37340077],
         [-21.02379604,   1.52321322],
         [-22.09072203,   1.48013644]],

        [[  1.        ,   1.        ],
         [ -1.41421356,   0.78539816],
         [  1.        ,   1.        ],
         ..., 
         [  4.12310563,   1.32581766],
         [  4.        ,   1.        ],
         [  4.12310563,   1.32581766]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  0.        ,   0.        ],
         [  1.        ,   1.        ],
         [ -2.23606798,   1.10714872],
         ..., 
         [ -3.16227766,   0.32175055],
         [  1.        ,   1.        ],
         [  1.41421356,  -0.78539816]],

        [[  1.        ,   1.        ],
         [  1.        ,   1.        ],
         [  1.        ,   1.        ],
         ..., 
         [  3.        ,   1.        ],
         [  2.        ,   1.        ],
         [ -1.41421356,   0.78539816]],

        ..., 
        [[  5.38516481,  -1.19028995],
         [  4.47213595,  -1.10714872],
         [  4.12310563,  -1.32581766],
         ..., 
         [  2.23606798,  -0.46364761],
         [  1.        ,   1.        ],
         [ -1.        ,  -0.        ]],

        [[ -5.38516481,   1.19028995],
         [ -4.12310563,   1.32581766],
         [ -3.16227766,   1.24904577],
         ..., 
         [  0.        ,   0.        ],
         [  1.        ,   0.        ],
         [  1.41421356,  -0.78539816]],

        [[  8.06225775,   1.44644133],
         [ -7.07106781,  -1.42889927],
         [  6.        ,   1.        ],
         ..., 
         [ -3.16227766,  -0.32175055],
         [ -3.16227766,  -0.32175055],
         [ -3.16227766,  -0.32175055]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[ -2.23606798,   0.46364761],
         [ -1.41421356,   0.78539816],
         [ -2.23606798,   0.46364761],
         ..., 
         [  1.        ,   0.        ],
         [  1.        ,   0.        ],
         [  1.        ,   1.        ]],

        [[ -2.23606798,  -0.46364761],
         [ -1.41421356,  -0.78539816],
         [  2.        ,   1.        ],
         ..., 
         [  0.        ,   0.        ],
         [  1.        ,   0.        ],
         [  1.        ,   0.        ]],

        ..., 
        [[  1.        ,   0.        ],
         [  1.        ,   1.        ],
         [ -2.23606798,  -1.10714872],
         ..., 
         [ 19.02629759,   1.51821327],
         [ 19.        ,   1.        ],
         [-19.10497317,  -1.46591939]],

        [[  3.60555128,   0.98279372],
         [  3.60555128,   0.5880026 ],
         [  5.        ,   0.64350111],
         ..., 
         [  7.28010989,  -1.29249667],
         [  7.61577311,  -1.16590454],
         [  8.06225775,  -1.05165021]],

        [[ -7.28010989,   1.29249667],
         [ -5.        ,   0.92729522],
         [ -5.83095189,   0.5404195 ],
         ..., 
         [ 20.09975124,   1.47112767],
         [ 21.02379604,   1.52321322],
         [-20.22374842,  -1.42190638]]]])

>> main
array([[  0.        ,   0.        ,   0.        , ...,  -0.78539816,
          2.23606798,  -1.10714872],
       [  0.        ,   0.        ,   0.        , ...,   1.        ,
         -2.23606798,  -1.10714872],
       [  0.        ,   0.        ,   0.        , ...,   1.        ,
          1.41421356,   0.78539816],
       ..., 
       [  0.        ,   0.        ,   0.        , ...,   1.        ,
          4.12310563,   1.32581766],
       [  0.        ,   0.        ,   0.        , ...,  -0.32175055,
         -3.16227766,  -0.32175055],
       [  0.        ,   0.        ,   0.        , ...,   1.52321322,
        -20.22374842,  -1.42190638]])
>> main.shape
(60, 2312)

方向和速度特征是非常有价值的（最重要的特征），因为它代表了每帧每个面部地标的运动，我正试图让机器学习算法在此基础上进行训练

我试图将三个维度重新塑造成一个长向量（只是将速度、方向和帧混合在一起），最后形成一个2D矩阵，我将其输入sklearn SVM函数，它产生的精度相当低。我预料到了这一点，因为我认为ml算法不可能识别出巨大的单个矩阵中的特征之间的差异，并假设向量中的所有内容都是相同的特征

我被迫制作2D矩阵，通过将速度、方向和每帧视频都强制加入一个向量，将其输入sklearn SVM，并获得低精度：

>> main.shape  
(60, 17, 68, 2)  
# 60 videos, 17 frames per video, 68 facial landmarks, 2 features (direction and speed)  
>> main  
array([[[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  1.        ,   1.        ],
         [  1.41421356,   0.78539816],
         [  1.41421356,   0.78539816],
         ..., 
         [  3.        ,   1.        ],
         [  3.        ,   1.        ],
         [  3.        ,   1.        ]],

        [[  0.        ,   0.        ],
         [ -1.41421356,   0.78539816],
         [ -1.41421356,   0.78539816],
         ..., 
         [  2.        ,   1.        ],
         [  3.        ,   1.        ],
         [  3.        ,   1.        ]],

        ..., 
        [[  1.        ,   1.        ],
         [  1.41421356,  -0.78539816],
         [  1.41421356,  -0.78539816],
         ..., 
         [ -1.41421356,   0.78539816],
         [  1.        ,   1.        ],
         [ -1.41421356,   0.78539816]],

        [[  2.23606798,  -0.46364761],
         [  2.82842712,  -0.78539816],
         [  2.23606798,  -0.46364761],
         ..., 
         [  1.        ,   0.        ],
         [  0.        ,   0.        ],
         [  1.        ,   1.        ]],

        [[ -1.41421356,  -0.78539816],
         [ -2.23606798,  -0.46364761],
         [ -2.23606798,  -0.46364761],
         ..., 
         [  1.41421356,  -0.78539816],
         [  1.41421356,  -0.78539816],
         [  2.23606798,  -1.10714872]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  2.        ,   1.        ],
         [  2.23606798,  -1.10714872],
         [  1.41421356,  -0.78539816],
         ..., 
         [ -2.        ,  -0.        ],
         [ -1.        ,  -0.        ],
         [ -1.41421356,  -0.78539816]],

        [[  2.        ,   1.        ],
         [ -2.23606798,   1.10714872],
         [ -1.41421356,   0.78539816],
         ..., 
         [  1.        ,   1.        ],
         [ -1.        ,  -0.        ],
         [ -1.        ,  -0.        ]],

        ..., 
        [[ -2.        ,  -0.        ],
         [ -3.        ,  -0.        ],
         [ -4.12310563,  -0.24497866],
         ..., 
         [  0.        ,   0.        ],
         [ -1.        ,  -0.        ],
         [ -2.23606798,   1.10714872]],

        [[ -2.23606798,   1.10714872],
         [ -1.41421356,   0.78539816],
         [ -2.23606798,   1.10714872],
         ..., 
         [ -2.23606798,   0.46364761],
         [ -1.41421356,   0.78539816],
         [ -1.41421356,   0.78539816]],

        [[  2.        ,   1.        ],
         [  1.41421356,   0.78539816],
         [  2.82842712,   0.78539816],
         ..., 
         [  1.        ,   1.        ],
         [  1.        ,   1.        ],
         [ -2.23606798,  -1.10714872]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  1.        ,   1.        ],
         [  0.        ,   0.        ],
         [  1.        ,   1.        ],
         ..., 
         [ -3.        ,  -0.        ],
         [ -2.        ,  -0.        ],
         [  0.        ,   0.        ]],

        [[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  1.41421356,   0.78539816],
         [  1.        ,   0.        ],
         [  0.        ,   0.        ]],

        ..., 
        [[  1.        ,   0.        ],
         [  1.        ,   1.        ],
         [  0.        ,   0.        ],
         ..., 
         [  2.        ,   1.        ],
         [  3.        ,   1.        ],
         [  3.        ,   1.        ]],

        [[ -7.28010989,   1.29249667],
         [ -7.28010989,   1.29249667],
         [ -8.54400375,   1.21202566],
         ..., 
         [-22.02271555,   1.52537305],
         [ 22.09072203,  -1.48013644],
         [ 22.36067977,  -1.39094283]],

        [[  1.        ,   0.        ],
         [  1.41421356,  -0.78539816],
         [  1.        ,   0.        ],
         ..., 
         [ -1.41421356,  -0.78539816],
         [  1.        ,   1.        ],
         [  1.41421356,   0.78539816]]],


       ..., 
       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  5.38516481,   0.38050638],
         [  5.09901951,   0.19739556],
         [  4.47213595,  -0.46364761],
         ..., 
         [ -1.41421356,   0.78539816],
         [ -2.82842712,   0.78539816],
         [ -5.        ,   0.64350111]],

        [[ -6.32455532,   0.32175055],
         [ -6.08276253,  -0.16514868],
         [ -5.65685425,  -0.78539816],
         ..., 
         [  3.60555128,   0.98279372],
         [  5.        ,   0.92729522],
         [  5.65685425,   0.78539816]],

        ..., 
        [[ -3.16227766,  -0.32175055],
         [ -3.60555128,  -0.98279372],
         [  5.        ,   1.        ],
         ..., 
         [ 12.08304597,   1.14416883],
         [ 13.15294644,   1.418147  ],
         [ 14.31782106,   1.35970299]],

        [[  3.60555128,  -0.5880026 ],
         [  4.47213595,  -1.10714872],
         [  6.        ,   1.        ],
         ..., 
         [-20.39607805,   1.37340077],
         [-21.02379604,   1.52321322],
         [-22.09072203,   1.48013644]],

        [[  1.        ,   1.        ],
         [ -1.41421356,   0.78539816],
         [  1.        ,   1.        ],
         ..., 
         [  4.12310563,   1.32581766],
         [  4.        ,   1.        ],
         [  4.12310563,   1.32581766]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[  0.        ,   0.        ],
         [  1.        ,   1.        ],
         [ -2.23606798,   1.10714872],
         ..., 
         [ -3.16227766,   0.32175055],
         [  1.        ,   1.        ],
         [  1.41421356,  -0.78539816]],

        [[  1.        ,   1.        ],
         [  1.        ,   1.        ],
         [  1.        ,   1.        ],
         ..., 
         [  3.        ,   1.        ],
         [  2.        ,   1.        ],
         [ -1.41421356,   0.78539816]],

        ..., 
        [[  5.38516481,  -1.19028995],
         [  4.47213595,  -1.10714872],
         [  4.12310563,  -1.32581766],
         ..., 
         [  2.23606798,  -0.46364761],
         [  1.        ,   1.        ],
         [ -1.        ,  -0.        ]],

        [[ -5.38516481,   1.19028995],
         [ -4.12310563,   1.32581766],
         [ -3.16227766,   1.24904577],
         ..., 
         [  0.        ,   0.        ],
         [  1.        ,   0.        ],
         [  1.41421356,  -0.78539816]],

        [[  8.06225775,   1.44644133],
         [ -7.07106781,  -1.42889927],
         [  6.        ,   1.        ],
         ..., 
         [ -3.16227766,  -0.32175055],
         [ -3.16227766,  -0.32175055],
         [ -3.16227766,  -0.32175055]]],


       [[[  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         ..., 
         [  0.        ,   0.        ],
         [  0.        ,   0.        ],
         [  0.        ,   0.        ]],

        [[ -2.23606798,   0.46364761],
         [ -1.41421356,   0.78539816],
         [ -2.23606798,   0.46364761],
         ..., 
         [  1.        ,   0.        ],
         [  1.        ,   0.        ],
         [  1.        ,   1.        ]],

        [[ -2.23606798,  -0.46364761],
         [ -1.41421356,  -0.78539816],
         [  2.        ,   1.        ],
         ..., 
         [  0.        ,   0.        ],
         [  1.        ,   0.        ],
         [  1.        ,   0.        ]],

        ..., 
        [[  1.        ,   0.        ],
         [  1.        ,   1.        ],
         [ -2.23606798,  -1.10714872],
         ..., 
         [ 19.02629759,   1.51821327],
         [ 19.        ,   1.        ],
         [-19.10497317,  -1.46591939]],

        [[  3.60555128,   0.98279372],
         [  3.60555128,   0.5880026 ],
         [  5.        ,   0.64350111],
         ..., 
         [  7.28010989,  -1.29249667],
         [  7.61577311,  -1.16590454],
         [  8.06225775,  -1.05165021]],

        [[ -7.28010989,   1.29249667],
         [ -5.        ,   0.92729522],
         [ -5.83095189,   0.5404195 ],
         ..., 
         [ 20.09975124,   1.47112767],
         [ 21.02379604,   1.52321322],
         [-20.22374842,  -1.42190638]]]])

>> main
array([[  0.        ,   0.        ,   0.        , ...,  -0.78539816,
          2.23606798,  -1.10714872],
       [  0.        ,   0.        ,   0.        , ...,   1.        ,
         -2.23606798,  -1.10714872],
       [  0.        ,   0.        ,   0.        , ...,   1.        ,
          1.41421356,   0.78539816],
       ..., 
       [  0.        ,   0.        ,   0.        , ...,   1.        ,
          4.12310563,   1.32581766],
       [  0.        ,   0.        ,   0.        , ...,  -0.32175055,
         -3.16227766,  -0.32175055],
       [  0.        ,   0.        ,   0.        , ...,   1.52321322,
        -20.22374842,  -1.42190638]])
>> main.shape
(60, 2312)

我想保留速度和方向特征，但必须在2D矩阵中表示它们，该矩阵考虑了视频中的帧

情感标签将贴在每个视频的17帧中的每一帧上。（因此，基本上，17帧视频将被标记为情感）

---

有什么聪明的方法可以重塑和缩小4D矩阵来实现这一点吗？

因此，你提出问题的方式绝对会让你看到精确度很低，你几乎无法改变它。将单一情感分配给视频（取决于您的语料库）通常不够准确，以至于任何机器学习算法都无法学习您试图提取的信号

此外，您将该问题定义为一个时间序列问题，这将使您的生活变得令人头痛，特别是如果您使用的是现成的

sklearn

算法，这些算法非常不适合此类任务

如果可能的话，您应该将您的问题定义为计算机视觉问题。你应该尝试预测每一帧的情绪内容。如果您没有具有这种粒度级别的数据集，您就不会看到很高的准确性

这有点偏离了你提问的方式，但你提问的方式是不易处理的。以下是解决问题的方法：

• 用情感内容标记单个框架
• 训练一个基于图像的算法来分类那些被标记的帧
  • 卷积神经网络可能会为任何基于图像的问题提供最佳性能，因为您有一个适当大小的数据集
  • 如果这不是一个选项，您需要开发图像的一维特征表示。我个人建议使用图像功能API。一旦你有了这个表示，一个典型的算法，比如SVM，将会非常有效
• 如果准确度不太符合您的要求，但越来越接近，我建议使用预处理/数据增强管道，如详细信息所示，该示例用于浮游生物识别，基本方法相同
• 如果准确度仍然达不到标准，并且您需要对整个视频进行预测，那么您将希望汇总结果，以在整个视频中给出准确的结果
  • 一种方法是在视频预测图上训练卷积神经网络。这有点奇怪，但可能效果很好
  • 一个好的方法是使用贝叶斯方法，假设每个预测都有一定程度的可信度，并结合视频上的预测分布
  • 最好的方法是将其视为集成学习问题。幸运的是，集成学习是一个被很好地研究和理解的问题。您可以找到如何以这种格式组合多个预测的详细信息

我希望这是有帮助的！如果你还有任何问题，请告诉我

---

免责声明：我是Indio的首席执行官，因此在推荐其使用时可能存在偏见。

您使用的是scikit learn and

numpy

。这不是一个MATLAB问题。您正在使用Python。因此，我重新标记了你的帖子以反映这一点。如果您的问题不涉及MATLAB环境，请避免使用MATLAB标签。谢谢您的回答！我将重新考虑我解决这个问题的方法，并尝试你的方法。对于功能，您是否仍然认为我在为每个帧添加情感标签以指示帧的上一个移动时仍保留速度和方向功能，或者您是否建议我删除这些功能并创建一组全新的功能？因此，添加速度和方向功能可能有助于提高准确性，但如果我不得不猜测，这种改进是微不足道的，我可能会采用神经网络方法。我想利用这一事实，我可以使用面部地标，并希望创建一个基于面部运动的情感识别系统。您会推荐什么样的功能来优化我的ac