validate.py
Validate optical flow estimation performance on standard datasets.
- validate.generate_outputs(args: Namespace, inputs: Dict[str, Tensor], preds: Dict[str, Tensor], dataloader_name: str, batch_idx: int, metadata: Dict[str, Any] | None = None) None[source]
Display on screen and/or save outputs to disk, if required.
- Parameters:
- argsNamespace
The arguments with the required values to manage the outputs.
- inputsDict[str, torch.Tensor]
The inputs loaded from the dataset (images, groundtruth).
- predsDict[str, torch.Tensor]
The model predictions (optical flow and others).
- dataloader_namestr
A string to identify from which dataloader these inputs came from.
- batch_idxint
Indicates in which position of the loader this input is.
- metadataDict[str, Any], optional
Metadata about this input, if available.
- validate.validate(args: Namespace, model: BaseModel) DataFrame[source]
Perform the validation.
- Parameters:
- argsNamespace
Arguments to configure the model and the validation.
- modelBaseModel
The model to be used for validation.
- Returns:
- pd.DataFrame
A DataFrame with the metric results.
See also
ptlflow.models.base_model.base_model.BaseModelThe parent class of the available models.
- validate.validate_list_of_models(args: Namespace) None[source]
Perform the validation.
- Parameters:
- argsNamespace
Arguments to configure the list of models and the validation.
- validate.validate_one_dataloader(args: Namespace, model: BaseModel, dataloader: DataLoader, dataloader_name: str) Dict[str, float][source]
Perform validation for all examples of one dataloader.
- Parameters:
- argsNamespace
Arguments to configure the model and the validation.
- modelBaseModel
The model to be used for validation.
- dataloaderDataLoader
The dataloader for the validation.
- dataloader_namestr
A string to identify this dataloader.
- Returns:
- Dict[str, float]
The average metric values for this dataloader.