使用Deepid的模型进行人脸表情识别
这几天因为前面的模型训练到70%左右便无法进行,因此该用deepid的模型改进后进识别,deepID增加了两层
layer {
name: “fc160”
type: “Eltwise”
bottom: “fc160_1”
bottom: “fc160_2”
top: “fc160”
eltwise_param {
operation: SUM
}
}
layer {
name: “dropout”
type: “Dropout”
bottom: “fc160”
top: “fc160”
dropout_param {
dropout_ratio: 0.4
}
}
训练准确率达到了92%左右,分别不同表情的效果得到了提高。
最后75000次迭代结果
75000 acurracy=0.91
75200 acurracy=0.92
75400 acurracy=0.85
75600 acurracy=0.88
75800 acurracy=0.92
76000 acurracy=0.89
76200 acurracy=0.9
76400 acurracy=0.93
76600 acurracy=0.88
76800 acurracy=0.93
77000 acurracy=0.86
77200 acurracy=0.88
77400 acurracy=0.86
77600 acurracy=0.92
77800 acurracy=0.87
78000 acurracy=0.9
78200 acurracy=0.91
78400 acurracy=0.93
78600 acurracy=0.92
78800 acurracy=0.91
79000 acurracy=0.89
79200 acurracy=0.92
79400 acurracy=0.91
79600 acurracy=0.9
79800 acurracy=0.86
80000 acurracy=0.93
80200 acurracy=0.95
80400 acurracy=0.92
80600 acurracy=0.88