vp_suite.measure.loss_provider

class PredictionLossProvider(config)

Bases: object

This class provides bundled access to multiple losses. With this class’s get_losses() method, all specified loss values are calculated on the same input prediction and target tensor.

device

A string specifying whether to use the GPU for calculations (cuda) or the CPU (cpu).

Type: str

losses

The concrete instantiated losses that the loss provider uses when provided with input tensors.

Type: dict

__init__(config)

Initializes the provider by extracting device and loss IDs from the provided config dict and instantiating the losses that shall be used.

Parameters: config (dict) – A dictionary containing the devices and losses to use. The provided losses come with the scales that should be multiplied by the respective loss value.

get_losses(pred, target)

Takes in tensors of predicted frames and the corresponding ground truth and calculates the losses for the loss classes instantiated previously. Each loss

Parameters

pred (torch.Tensor) – The predicted frame sequence as a 5D float tensor (batch, frames, c, h, w).
target (torch.Tensor) – The ground truth frame sequence as a 5D float tensor (batch, frames, c, h, w)

Returns

A dictionary containing the loss IDs and the corresponding values of each loss in display representation.
The sum of scaled losses (total loss).