【Loss】Centerloss代码详解（pytorch）完整版.pdf

【Loss】Centerloss代码详解 （pytorch） **注 ：**全部代码在最后，此代码不知来 ⾃哪位⼤神。 m:ba ch size n:class size d:fea dim ⎛ x0 ⎞ ⎛ x00 x0 1 ... x0 (d−1) ⎞ ⎜ x1 ⎟ ⎜ x10 x11 ... x1(d−1) ⎟ ⎜ ⎟ ⎜ ⎟ ⎝ ... ⎠ ⎝ ... ... ... ⎠ x = xm−1 = x(m−1)0 x(m−1)1 ... x(m−1)(d−1) ∈ Rm ×d torch.pow(x, 2).sum(dim=1, keepdim=True) 2 2 2 2 x00 + x0 1 + ... + x0 (d−1) ⎛ x0 ⎞ ⎛ 2 2 2 ⎞ 2 x10 + x11 + ... + x1(d−1) ⎜ x1 ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ... ... 2 2 2 2 ⎝ ⎠ ⎝ (m−1)0 (m−1)1 (m−1)(d−1)⎠ = xm−1 = x + x + ... + x ∈ Rm ×1 .expand(batch size, self.num classes) 2 2 2 ⎛ x0 x0 ... x0 ⎞ 2 2 2 ⎜ x1 x1 ... x1 ⎟ ⎜ ⎟ ... ... ... ... ⎝ 2 2 2 ⎠ m ×n = xm−1 xm−1 ... xm−1 ∈ R 同理 self.centers = nn.Parameter(torch.randn(self.num classes, self.feat dim).cuda()) ⎛ c00 c0 1 ... c0 (d−1) ⎞ ⎜ c10 c11 ... c1(d−1) ⎟ ⎜ ⎟ ⎝ ... ... ... ⎠ centers = c(n−1)0 c(n−1)1 ... c(n−1)(d−1) ∈ Rn ×d torch.pow(self.centers, 2).sum(dim=1, keepdim=True) 2