Called during inception. Don't encode. sample from N(0, 1) :param V: :param time_len: :param seed: :return:

Inputs: - vecs: Tensor of shape (O, D) giving vectors - boxes: Tensor of shape (O, 4) giving bounding boxes in the format [xc, yc,

:param edges: :param bbox: V, O, 2 :return:

Convert boxes from [xc, yc, w, h] format to [x0, y0, x1, y1] format Input: - boxes: Input boxes of shape (N, 4) in [xc, yc, w, h] format

merge_net(bg_net(bg)), mask with obj(obj_net)

:param u: (1, V, D) :return: (dt, V, D)

:param u: (1, V, D) :return: (dt, V, D)

Compute the box transform that, when applied to anchors, would give boxes. Inputs: - anchors: Box anchors of shape (N, 4) in the format

Called during realistic testing. 1. encode u. 2. NO sample :return: obj_z: (Dt, V, D), kl_loss: criterion

generate sample from N(mu, var). or no sample

:param u: (dt - 1, V, D) :return: (dt, V, D). random generalize one

:param u: (dt - 1, V, D) :return: (dt, V, D). random generalize one

:param u: (1, V, D) :return: (1, V, D). random generalize one

error: (O, ). Update total-dt, oi-dt...

:return: src, dst. each is a list of object - 'image': FloatTensor of shape (dt, C, H, W). resize and normalize to faster-rcnn

:return: src, dst. each is a list of object - 'image': FloatTensor of shape (dt, C, H, W). resize and normalize to faster-rcnn

:param dims: channel dimensions

:param dims: channel dimensions

under construction!!!

under construction!!!

under construction!!!