WandbVisBackend¶

class mmengine.visualization.WandbVisBackend(save_dir, init_kwargs=None, define_metric_cfg=None, commit=True, log_code_name=None, watch_kwargs=None)[source]¶

Wandb visualization backend class.

Examples

>>> from mmengine.visualization import WandbVisBackend
>>> import numpy as np
>>> wandb_vis_backend = WandbVisBackend()
>>> img=np.random.randint(0, 256, size=(10, 10, 3))
>>> wandb_vis_backend.add_image('img', img)
>>> wandb_vis_backend.add_scaler('mAP', 0.6)
>>> wandb_vis_backend.add_scalars({'loss': [1, 2, 3],'acc': 0.8})
>>> cfg = Config(dict(a=1, b=dict(b1=[0, 1])))
>>> wandb_vis_backend.add_config(cfg)

Parameters:

save_dir (str, optional) – The root directory to save the files produced by the visualizer.
init_kwargs (dict, optional) – wandb initialization input parameters. See wandb.init for details. Defaults to None.
define_metric_cfg (dict or list[dict], optional) – When a dict is set, it is a dict of metrics and summary for wandb.define_metric. The key is metric and the value is summary. When a list is set, each dict should be a valid argument of the define_metric. For example, define_metric_cfg={'coco/bbox_mAP': 'max'}, means the maximum value of coco/bbox_mAP is logged on wandb UI. When define_metric_cfg=[dict(name='loss', step_metric='epoch')], the “loss” will be plotted against the epoch. See wandb define_metric for details. Defaults to None.
commit (bool, optional) – and increment the step. If false wandb.log just updates the current metrics dict with the row argument and metrics won’t be saved until wandb.log is called with commit=True. Defaults to True.
log_code_name (str, optional) – By default, the artifact will be named source-$PROJECT_ID-$ENTRYPOINT_RELPATH. See wandb log_code for details. Defaults to None. New in version 0.3.0.
watch_kwargs (optional, dict) – Agurments for wandb.watch. New in version 0.4.0.

add_config(config, **kwargs)[source]¶

Record the config to wandb.

Parameters:: config (Config) – The Config object
Return type:: None

add_graph(model, data_batch, **kwargs)[source]¶

Record the model graph.

Parameters:

model (torch.nn.Module) – Model to draw.
data_batch (Sequence[dict]) – Batch of data from dataloader.

Return type:

None

add_image(name, image, step=0, **kwargs)[source]¶

Record the image to wandb.

Parameters:

name (str) – The image identifier.
image (np.ndarray) – The image to be saved. The format should be RGB.
step (int) – Useless parameter. Wandb does not need this parameter. Defaults to 0.

Return type:

None

add_scalar(name, value, step=0, **kwargs)[source]¶

Record the scalar data to wandb.

Parameters:

name (str) – The scalar identifier.
value (int, float, torch.Tensor, np.ndarray) – Value to save.
step (int) – Useless parameter. Wandb does not need this parameter. Defaults to 0.

Return type:

None

add_scalars(scalar_dict, step=0, file_path=None, **kwargs)[source]¶

Record the scalar’s data to wandb.

Parameters:

scalar_dict (dict) – Key-value pair storing the tag and corresponding values.
step (int) – Useless parameter. Wandb does not need this parameter. Defaults to 0.
file_path (str, optional) – Useless parameter. Just for interface unification. Defaults to None.

Return type:

None

close()[source]¶

close an opened wandb object.

Return type:: None

property experiment¶

Return wandb object.

The experiment attribute can get the wandb backend, If you want to write other data, such as writing a table, you can directly get the wandb backend through experiment.