denormalize_box_params in the norm_tensor

[caixiaoniweimar]

Hi,

I have a question regarding obj_pointset = self.norm_tensor(obj_pointset, denormalize_box_params(tight_boxes[i]) Could you please help me out?

It seems like you assigned a number from 0 to 24 to angle(bbox[6]), right? and then you append the box to the tight_boxes.

if not self.vae_baseline:
    bins = np.linspace(0, np.deg2rad(360), 24)
    angle = np.digitize(bbox[6], bins)
    bbox = normalize_box_params(bbox)
    bbox[6] = angle
else:
    bins = np.linspace(0, np.deg2rad(360), 24)
    bbox[6] = np.digitize(bbox[6], bins)
tight_boxes.append(bbox)

However, when you try to denormalize the box parameters before passing to the norm_tensor method, the statistics you used in the denormalize method is a radius number instead of a number from 0 to 24,right?

if not self.vae_baseline:
    obj_pointset = self.norm_tensor(obj_pointset, denormalize_box_params(tight_boxes[i]),
                           scale=True, rotation=self.use_canonical, scale_func=self.scale_func)

It doesn't make much sense to me at the moment. Why did you do that?

Thanks for your help!

Regards,