Temporal amorized version

joint reconstruction Serves as the DetModel, handles fusion/communication between encoder and decoder

Temporal amorized version

wrapper of vqvae, handles fusion

Modified from: https://arxiv.org/pdf/2012.12395.pdf Args: config (object): The Config object. layer (int, optional): Col

3D cnn used in the encoder.

a upsampling and deconve module in repalce of the original final linear projection, to obtain the prediction of the original image. for p

a upsampling and deconve module in repalce of the original final linear projection, to obtain the prediction of the original image. for p

Track a series of values and provide access to smoothed values over a window or the global series average.

individual reconstruction Serves as the DetModel, handles fusion/communication after encoder and decoder

Masked Autoencoder with VisionTransformer backbone

Used by non-intermediate models. Pass the output from encoder directly to decoder.

STAR + Vector Quantizer

Use vqvae to reconstruct the scene

The backbone class that contains encode and decode function