CurricularFace[2020-CVPR]

Prior works

• MV-Softmax[2020-AAAI] [强烈建议]

Motivation

• MV-Sotamax存在的问题：从training起始阶段就开始强调semi-hard/hard-sample，可能会导致模型的收敛问题！      

          

• insight : easy sample first, hard sample later!       

Code

• CurricularFace[Pytorchvision]

class CurricularFace(nn.Module):
    def __init__(self, in_features, out_features, m = 0.5, s = 64.):
        super(CurricularFace, self).__init__()
        self.in_features = in_features
        self.out_features = out_features
        self.m = m
        self.s = s
        self.cos_m = math.cos(m)
        self.sin_m = math.sin(m)
        self.threshold = math.cos(math.pi - m)
        self.mm = math.sin(math.pi - m) * m
        self.kernel = Parameter(torch.Tensor(in_features, out_features))
        self.register_buffer('t', torch.zeros(1))
        nn.init.normal_(self.kernel, std=0.01)

    def forward(self, embbedings, label):
        embbedings = l2_norm(embbedings, axis = 1)
        kernel_norm = l2_norm(self.kernel, axis = 0)
        cos_theta = torch.mm(embbedings, kernel_norm)
        cos_theta = cos_theta.clamp(-1, 1)  # for numerical stability
        with torch.no_grad():
            origin_cos = cos_theta.clone()
        target_logit = cos_theta[torch.arange(0, embbedings.size(0)), label].view(-1, 1)

        sin_theta = torch.sqrt(1.0 - torch.pow(target_logit, 2))
        cos_theta_m = target_logit * self.cos_m - sin_theta * self.sin_m #cos(target+margin)
        mask = cos_theta > cos_theta_m
        final_target_logit = torch.where(target_logit > self.threshold, cos_theta_m, target_logit - self.mm)

        hard_example = cos_theta[mask]
        with torch.no_grad():
            self.t = target_logit.mean() * 0.01 + (1 - 0.01) * self.t
        cos_theta[mask] = hard_example * (self.t + hard_example)
        cos_theta.scatter_(1, label.view(-1, 1).long(), final_target_logit)
        output = cos_theta * self.s
        return output, origin_cos * self.s

Details

• Curricular Loss                                               

                                                                 

        其中，T(cos(θ_y)) = cos(θ_y + m), I (t, cos(θ_j))表示样本的权重函数，N(t, cos(θ_j))定义如下：

                                                   

• Training Curve

            

1. x-axis : iterations, y-axis : 难样本的调整系数[modulation coefficients];
2. t : adaptive parameter; M(MV-Arc-Softmax) : MV-Arc-Softmax; M(ours) : gradient modulation coefficients;                       
3. 在训练早期，t --> 0, I(t, cos(θ_j)) = 1,模型可以利用easy-sample加速收敛；在训练中后期t不断增大使得I(t, cos(θ_j)) > 1，这样模型可以更多地关注hard-smaples.

            eary,  later     

       Note : (a, b), a表示在训练过程中[某个时刻] curricular_loss和arcface-loss的比值；b表示max {cos(θ_j), j ≠ yi} 

                                           

• Adaptive Estimation of t

1. r^(k)表示第k个mini-batch中positive-cosin similarity的均值，r^(0) = 0;
2. , α = 0.99.  【大家可以脑补一下：为什么t^(k)随着k的增加，会呈现出单调递增的趋势呢？】
3. different strategies or vaule of t

                                                                

Experiment

• Benchmark

                                                            

• Challenge

                                                                                        

Reference

[1]. CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition[2020-CVPR]