Table 2.

Channels 5–16 of the image-space representation used to represent pig locations and orientations. Pairs of neighboring channels corresponds to the x and y offset between neighboring parts. Overall, these 12 channels represent bidirectional vectors linking three pairs of body parts.

Channel	5	6	7	8	9	10
Encoding	$\propto (x_{{\mathbf{l}}_n}-x_{{\mathbf{s}}_n})$	$\propto (y_{{\mathbf{l}}_n}-y_{{\mathbf{s}}_n})$	$\propto (x_{{\mathbf{s}}_n}-x_{{\mathbf{l}}_n})$	$\propto (y_{{\mathbf{s}}_n}-y_{{\mathbf{l}}_n})$	$\propto (x_{{\mathbf{r}}_n}-x_{{\mathbf{s}}_n})$	$\propto (y_{{\mathbf{r}}_n}-y_{{\mathbf{s}}_n})$
Channel	11	12	13	14	15	16
Encoding	$\propto (x_{{\mathbf{s}}_n}-x_{{\mathbf{r}}_n})$	$\propto (y_{{\mathbf{s}}_n}-y_{{\mathbf{r}}_n})$	$\propto (x_{{\mathbf{s}}_n}-x_{{\mathbf{t}}_n})$	$\propto (y_{{\mathbf{s}}_n}-y_{{\mathbf{t}}_n})$	$\propto (x_{{\mathbf{t}}_n}-x_{{\mathbf{s}}_n})$	$\propto (y_{{\mathbf{t}}_n}-y_{{\mathbf{s}}_n})$