How to visualize offset vectors?

[gymoon10]

As shown in figure 2, I want to visualize predicted offset vectors. But, there are no descriptions about how the offset vectors were visualized. I'm studying these papers and codes carefully, and I would really appreciate your help.

[Tgnink]

Tthe center has offsets vector too. But luckily, it seems that all objects' centroids have similar offsets. so you can show result like this:

feat0= output[0][0].cpu().numpy()
feat1 = output[0][1].cpu().numpy()
seed=  output[0][-1]
seed=torch.sigmoid(seed).cpu().numpy()

feat0=normalize(feat0,1,99.8)-0.5
feat1=normalize(feat1,1,99.8)-0.5
ang=np.angle(feat0 + (feat1) * 1j,deg=True)


hue=(ang+180)/360
satuation=seed[2]
value=seed[2]
hsv=np.stack([hue,satuation,value],axis=-1)
rgb=hsv_to_rgb(hsv)

the angle or RGB maybe is what you want.