LoggerHook visualization output doesn't provide per class validation values.

[Jonas-Bartkowski]

After training, my work_dir/vis_data/scalars.json looks like this:

{"lr": 1.2123528399645335e-06, "data_time": 0.015496420860290527, "loss": 0.9026781380176544, "decode.loss_ce": 0.6495033919811248, "decode.acc_seg": 34.280513763427734, "aux.loss_ce": 0.2531747445464134, "aux.acc_seg": 32.2140007019043, "time": 0.5931255578994751, "memory": 1 ...
{"aAcc": 24.2, "mIoU": 13.59, "mAcc": 53.12, "data_time": 0.007318027599437817, "time": 0.17899566727715568, "step": 20}
{"lr": 1.7166508700338139, ...}

I'm really missing the class labels here. It's very important to me that I can also see the per class validation output in scalars.json and tensorflow.

My logger and visualization hooks:

default_hooks = dict(
    # print log every 50 iterations.
    logger=dict(type='LoggerHook', interval=10, log_metric_by_epoch=False),

    # save checkpoint every 500 iterations.
    checkpoint=dict(type='CheckpointHook', by_epoch=False, interval=cp_period),
)

# TENSORBOARD HOOK

vis_backends = [dict(type='LocalVisBackend'), dict(type='TensorboardVisBackend')]
visualizer = dict(
    type='SegLocalVisualizer', vis_backends=vis_backends, name='visualizer', classes=classes, palette=[[128, 128, 128],[128, 255, 128]]
)

I noticed that the visualizer is not called in after_val_iter() in mmengine.hooks.logger_hook.py, like it is in after_val_epoch().
Is that that the source? If so, is that a bug?

def after_val_iter(self,
                       runner,
                       batch_idx: int,
                       data_batch: DATA_BATCH = None,
                       outputs: Optional[Sequence] = None) -> None:
        """Record logs after validation iteration.
        Args:
            runner (Runner): The runner of the validation process.
            batch_idx (int): The index of the current batch in the validation
                loop.
            data_batch (dict or tuple or list, optional): Data from dataloader.
                Defaults to None.
            outputs (sequence, optional): Outputs from model.
        """
        if self.every_n_inner_iters(batch_idx, self.interval):
            _, log_str = runner.log_processor.get_log_after_iter(
                runner, batch_idx, 'val')
            runner.logger.info(log_str)

If not, is there any other way to log validation results for individual classes into vis_data?

Help is much appreciated, thanks for this awesome framework. 🙏