fruit_project.utils.logging

Functions

initwandb(→ wandb.sdk.wandb_run.Run)	Initializes a wandb run.
get_run_name(→ str)	Generates a run name based on the configuration.
log_images(, mean, std)	Logs a grid of images with their bounding boxes to wandb.
log_transforms(→ None)	Logs a grid of transformed images with their bounding boxes to wandb.
log_training_time(→ None)	Logs the elapsed training time.
log_model_params(→ None)	Logs the total and trainable parameters of a model.
log_class_value_counts(→ None)	Logs the class distribution of a dataset.
log_checkpoint_artifact(→ None)	Logs a model checkpoint as a wandb artifact.
log_detection_confusion_matrix(→ None)	Logs a detection confusion matrix plot to wandb.
log_per_class_metric(→ None)
log_test_data(→ None)	Performs final validation, logs metrics, and logs it to wandb
log_epoch_data(→ None)	Constructs and logs a dictionary of metrics for logging at the end of an epoch.

Module Contents

fruit_project.utils.logging.initwandb(cfg: omegaconf.DictConfig) → wandb.sdk.wandb_run.Run

Initializes a wandb run.

Parameters:

cfg (DictConfig) – Configuration object.

Returns:

The wandb run object.

Return type:

Run

fruit_project.utils.logging.get_run_name(cfg: omegaconf.DictConfig) → str

Generates a run name based on the configuration.

Parameters:

cfg (DictConfig) – Configuration object.

Returns:

The generated run name.

Return type:

str

fruit_project.utils.logging.log_images(run: wandb.sdk.wandb_run.Run, batch: Dict, id2lbl: Dict, grid_size: Tuple[int, int] = (3, 3), mean: torch.Tensor | None = None, std: torch.Tensor | None = None) → None

Logs a grid of images with their bounding boxes to wandb.

Parameters:

• run (Run) – The wandb run object.

• batch (Tuple[Dict, List]) – A single batch of data (processed_batch, targets).

• id2lbl (Dict) – A dictionary mapping class IDs to labels.

• grid_size (Tuple[int, int], optional) – The grid size for displaying images. Defaults to (3, 3).

• mean (Optional[torch.Tensor], optional) – The mean used for normalization. Defaults to None.

• std (Optional[torch.Tensor], optional) – The standard deviation used for normalization. Defaults to None.

fruit_project.utils.logging.log_transforms(run: wandb.sdk.wandb_run.Run, batch: Dict, grid_size: Tuple[int, int], id2lbl: Dict[int, str], transforms: Dict | None = None, mean: torch.Tensor | None = None, std: torch.Tensor | None = None, do_normalize: bool = False) → None

Logs a grid of transformed images with their bounding boxes to wandb.

Parameters:

• run (Run) – The wandb run object.

• batch (Tuple[Dict, List]) – A single batch of data (processed_batch, targets).

• grid_size (Tuple[int, int], optional) – The grid size for displaying images. Defaults to (3, 3).

• id2lbl (Optional[Dict], optional) – A dictionary mapping class IDs to labels. Defaults to None.

• transforms (Optional[Dict], optional) – The transforms applied. Defaults to None.

• mean (Optional[torch.Tensor], optional) – The mean used for normalization. Defaults to None.

• std (Optional[torch.Tensor], optional) – The standard deviation used for normalization. Defaults to None.

fruit_project.utils.logging.log_training_time(run: wandb.sdk.wandb_run.Run, start_time: float) → None

Logs the elapsed training time.

Parameters:

• run (Run) – The wandb run object.

• start_time (float) – The start time of training.

fruit_project.utils.logging.log_model_params(run: wandb.sdk.wandb_run.Run, model: torch.nn.Module) → None

Logs the total and trainable parameters of a model.

Parameters:

• run (Run) – The wandb run object.

• model (nn.Module) – The model.

fruit_project.utils.logging.log_class_value_counts(run: wandb.sdk.wandb_run.Run, samples: List[Tuple[str, str]], stage: str = 'Train') → None

Logs the class distribution of a dataset.

Parameters:

• run (Run) – The wandb run object.

• samples (List[Tuple[Any, Any]]) – A list of samples (e.g., [(image, label), …]).

• stage (str, optional) – The dataset stage (e.g., ‘Train’, ‘Test’). Defaults to “Train”.

fruit_project.utils.logging.log_checkpoint_artifact(run: wandb.sdk.wandb_run.Run, path: str, name: str, epoch: int, wait: bool = False) → None

Logs a model checkpoint as a wandb artifact.

Parameters:

• run (Run) – The wandb run object.

• path (str) – The path to the checkpoint file.

• name (str) – The name of the artifact.

• epoch (int) – The epoch number.

• wait (bool, optional) – Whether to wait for the artifact to be uploaded. Defaults to False.

fruit_project.utils.logging.log_detection_confusion_matrix(run: wandb.sdk.wandb_run.Run, cm_object: fruit_project.utils.metrics.ConfusionMatrix, class_names: List[str]) → None

Logs a detection confusion matrix plot to wandb.

Parameters:

• run (Run) – The wandb run object.

• cm_object (ConfusionMatrix) – The confusion matrix object.

• class_names (List[str]) – The list of class names.

fruit_project.utils.logging.log_per_class_metric(class_names: List, metric_per_class: torch.Tensor, ds_type: str, metric_name: str, log_data: Dict) → None

fruit_project.utils.logging.log_test_data(epoch: int, best_val_map: float, trainer) → None

Performs final validation, logs metrics, and logs it to wandb

Parameters:

• epoch (int) – The final epoch number.

• best_test_map (float) – The best test mAP@.50 achieved during training.

Returns:

None

fruit_project.utils.logging.log_epoch_data(epoch: int, train_loss: Dict[str, float], val_loss: Dict[str, float], val_metrics: Dict, trainer) → None

Constructs and logs a dictionary of metrics for logging at the end of an epoch.

Parameters:

• epoch (int) – The current epoch number.

• train_loss (Dict) – Dict containing total, classification, bbox and giou training loss for epoch.

• test_map (float) – The test mAP@.5-.95.

• test_map50 (float) – The test mAP@.50.

• test_loss (Dict) – Dict containing total, classification, bbox and giou test loss for epoch.

• test_map_per_class (torch.Tensor) – The test mAP@.50 for each class.

• trainer (Trainer) – instance of Trainer class

Returns:

None