基于光流和卷积视觉Transformer的轻量级微表情识别

徐恺蔚,KHIZER BIN TALIBHafiz,曹衍龙,许源平,许志杰,宋景春

Lightweight micro-expression recognition based on optical flow and convolutional vision Transformer

Kaiwei XU,Hafiz KHIZER BIN TALIB,Yanlong CAO,Yuanping XU,Zhijie XU,Jingchun SONG

表 5 不同模型输入的对比实验结果

Tab.5 Comparative experimental results of different model inputs

模型输入类型 时间帧 Full SMIC CASME Ⅱ SAMM UF1 UAR UF1 UAR UF1 UAR UF1 UAR RGB图像 峰值帧 0.761 9 0.758 1 0.648 7 0.647 0 0.903 4 0.893 6 0.743 5 0.727 4 光流 峰值帧 0.887 0 0.885 4 0.820 7 0.822 7 0.965 1 0.962 0 0.888 9 0.873 5 光流+光应变 均匀4帧 0.841 1 0.831 7 0.803 6 0.804 7 0.832 4 0.804 4 0.879 3 0.871 6 均匀2帧 0.853 7 0.857 5 0.825 2 0.833 3 0.845 7 0.850 0 0.890 6 0.867 8 峰值帧 0.910 2 0.910 2 0.851 2 0.854 7 0.985 8 0.985 8 0.908 0 0.893 6