data

VJEPA2ACOutput

extendsModelOutput

VJEPA2ACOutput(prediction: Tensor, context: Tensor, target: Tensor, loss: Tensor)

Latent transition output of V-JEPA 2-AC.

predictionTensor

Predicted latent grids, (B, S * P, D).

contextTensor

Encoded context grids consumed by the predictor.

targetTensor

The latents actually observed at the predicted frames.

lossTensor

Representation-space L1 between the two.

Constructors

dunder

→None

__init__(prediction: Tensor, context: Tensor, target: Tensor, loss: Tensor)