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 2 所提方法与现有方法的性能对比

Tab.2 Performance comparison of proposed method and existing methods

方法 Full SMIC CASME Ⅱ SAMM UF1 UAR UF1 UAR UF1 UAR UF1 UAR LBP-TOP[3] 0.5882 0.5785 0.2000 0.5280 0.7026 0.7429 0.3954 0.4102 Bi-WOOF[7] 0.6296 0.6227 0.5727 0.5829 0.7805 0.8026 0.5211 0.5139 STSTNet[1] 0.7353 0.7605 0.6801 0.7013 0.8382 0.8686 0.6588 0.6810 IncepTR[23] 0.7530 0.7460 0.6550 0.6500 0.9110 0.8960 0.6910 0.6940 EDSMISEViTNet[27] 0.7587 0.7736 0.7372 0.7139 0.8521 0.8461 0.7216 0.6781 SLSTT-LSTM[18] 0.8160 0.7900 0.7400 0.7200 0.9010 0.8850 0.7150 0.6430 BDCNN[15] 0.8509 0.8500 0.7859 0.7869 0.9501 0.9516 0.8186 0.7994 ViT-16/B[21] 0.8512 0.8397 0.8044 0.7994 0.9220 0.9137 0.8142 0.7847 HTNet[22] 0.8603 0.8475 0.8049 0.7905 0.9532 0.9516 0.8131 0.8124 MTMNet[10] 0.8640 0.8570 0.8640 0.8610 0.8700 0.8720 0.8250 0.8190 ResNet-18[12] 0.8696 0.8724 0.7918 0.7982 0.9594 0.9612 0.8820 0.8549 MiMaNet[11] 0.8830 0.8760 0.8730 0.8670 0.8810 0.8810 0.8960 0.8840 Micron-BERT[29] 0.8903 0.8842 — — — — — — MiER-CvT 0.9102 0.9102 0.8512 0.8547 0.9858 0.9858 0.9080 0.8936 MiER-CvT+身份域 0.9171 0.9192 0.8546 0.8601 0.9928 0.9896 0.9177 0.9064