What does vector w denote?

[FedericaGielen]

What does vector w denote?

[FedericaGielen]

What is b IN YOUR PAPER?

[YihongSun]

• $\vec{w}$ stands for the latent variables { $w_a$ }, which indicate foreground ($w_a=1$) or background ($w_a=0$) for each lattice position $a \in D$. Intuitively, $P(\vec{w} | y, m)$ is the probability of each lattice position $a \in D$ being a foreground pixel given the label $y$ and mixture component $m$, which is further discussed in the Shape Modeling paragraph.

  • Similarly, $\vec{z}$ stands for the latent variables { $z_a$ }, which indicate visible ($z_a=1$) or occluded ($z_a=0$) for each lattice position $a \in D$.

• $b$ stands for the given bounding box during training. Here, $L_{seg}$ adopts the MIL loss proposed in [16].