Hi @<1576381444509405184:profile|ManiacalLizard2> ! Can you please share a code snippet that I could run to investigate the issue?

Sure:

def main():
    repo = "redacted"
    commit = "redacted"
    commit = "redacted"
    bands = ["redacted"]
    test_size = 0.2
    batch_size = 64
    num_workers = 12
    img_size = (128, 128)
    random_seed = 42
    epoch = 20
    learning_rate = 0.1

    livbatch_list = get_livbatch_list(repo, commit)
    lbs = download_batches(repo, commit, livbatch_list)

    df, label_map = get_annotation_df(lbs, bands)

    df_train, df_val = deterministic_train_val(df, test_size=test_size)

    train_dataloader = dict(
        batch_size=batch_size,
        num_workers=num_workers,
        persistent_workers=True,
        sampler=dict(type='DefaultSampler', shuffle=True),
        dataset=LivDataset(
            anno_df=df_train,
            bands=bands,
            pipeline=[LivImageLoader(bands=bands, img_size=img_size),
                      Albumentations([
                          dict(
                              type='ShiftScaleRotate',
                              shift_limit=0.0625,
                              scale_limit=0.2,
                              rotate_limit=15,
                              interpolation=1,
                              border_mode=cv2.BORDER_CONSTANT,
                              p=0.5),
                      ]),
                      dict(type='PackInputs')]
        ),
    )

    val_dataloader = dict(
        batch_size=batch_size,
        num_workers=num_workers,
        persistent_workers=True,
        sampler=dict(type='DefaultSampler', shuffle=False),
        dataset=LivDataset(
            df_val,
            bands,
            pipeline=[LivImageLoader(bands=bands, img_size=img_size),
                      dict(type='PackInputs')]
        )
    )

    model = redacted
    
    val_evaluator = [
        dict(type='Accuracy', topk=(1, ), prefix='val/accuracy'),
        dict(
            type='SingleLabelMetric',
            items=('precision', 'recall', 'f1-score'),
            average='macro',
            prefix='val/macro',
        ),  # class-wise mean
        dict(
            type='SingleLabelMetric',
            items=('precision', 'recall', 'f1-score'),
            average='micro',
            prefix='val/micro',
        ),  # overall mean
    ]

    runner = Runner(
        # the model used for training and validation.
        # Needs to meet specific interface requirements
        default_scope="mmpretrain",
        model=model,
        # working directory which saves training logs and weight files
        work_dir='./work_dir',

        # train dataloader needs to meet the PyTorch data loader protocol
        train_dataloader=train_dataloader,

        # optimize wrapper for optimization with additional features like
        # AMP, gradtient accumulation, etc
        # optim_wrapper=dict(optimizer=dict(type=SGD, lr=0.001, momentum=0.9)),
        optim_wrapper=dict(type='OptimWrapper',
                           #    accumulative_counts=accumulation,
                           optimizer=dict(type='SGD',
                                          lr=learning_rate,
                                          momentum=0.9,
                                          weight_decay=0.0005),
                           clip_grad=None),

        # trainging coinfs for specifying training epoches, verification intervals, etc
        train_cfg=dict(by_epoch=True, max_epochs=epoch, val_interval=1),

        # validation dataloaer also needs to meet the PyTorch data loader protocol
        val_dataloader=val_dataloader,

        # validation configs for specifying additional parameters required for validation
        val_cfg=dict(),

        # validation evaluator. The default one is used here
        # val_evaluator=dict(type=Accuracy),
        val_evaluator=val_evaluator,

        randomness=dict(seed=random_seed),

        visualizer=dict(type='Visualizer', vis_backends=[dict(type='ClearMLVisBackend')]),

        default_hooks=dict(checkpoint=dict(
            type='CheckpointHook',
            max_keep_ckpts=1,
            save_last=True,
            save_best=[
                'auto',
            ],
        ),)
    )

    runner.train()

if __name__ == "__main__":
    task = clearml.Task.init(project_name="mmpretrain", task_name='mmpretrain')
    main()

Looks like it's because I did not do the mmpretrain way with dict config file ...