WZMIAOMIAO/deep-learning-for-image-processing · error · AttributeError

'{}' object has no attribute '{}'

Error message

'{}' object has no attribute '{}'

What it means

MetricLogger.__getattr__ in unet/train_utils raises AttributeError when the requested attribute is neither a registered meter (in self.meters) nor an instance attribute. It's a deliberate fail-loud check so typos on the metric logger surface immediately.

Source

Thrown at pytorch_segmentation/unet/train_utils/distributed_utils.py:187

class MetricLogger(object):
    def __init__(self, delimiter="\t"):
        self.meters = defaultdict(SmoothedValue)
        self.delimiter = delimiter

    def update(self, **kwargs):
        for k, v in kwargs.items():
            if isinstance(v, torch.Tensor):
                v = v.item()
            assert isinstance(v, (float, int))
            self.meters[k].update(v)

    def __getattr__(self, attr):
        if attr in self.meters:
            return self.meters[attr]
        if attr in self.__dict__:
            return self.__dict__[attr]
        raise AttributeError("'{}' object has no attribute '{}'".format(
            type(self).__name__, attr))

    def __str__(self):
        loss_str = []
        for name, meter in self.meters.items():
            loss_str.append(
                "{}: {}".format(name, str(meter))
            )
        return self.delimiter.join(loss_str)

    def synchronize_between_processes(self):
        for meter in self.meters.values():
            meter.synchronize_between_processes()

    def add_meter(self, name, meter):
        self.meters[name] = meter

    def log_every(self, iterable, print_freq, header=None):

View on GitHub (pinned to 1ec3fe6f37)

Solutions

  1. Correct the attribute name to a registered meter (loss, lr, ...)
  2. Ensure the meter is assigned before access: metric_logger.xxx = SmoothedValue()
  3. Use hasattr(metric_logger, name) or name in metric_logger.meters before reading

Example fix

// before
print(metric_logger.acc.global_avg)
// after
print(metric_logger.loss.global_avg)  # 'acc' was never registered on MetricLogger
Defensive patterns

Strategy: validation

Validate before calling

required = ['loss', 'lr']
missing = [r for r in required if r not in metric_logger.meters]
assert not missing, f"unregistered meters: {missing}"

Type guard

def meter_exists(logger, name: str) -> bool:
    return name in logger.meters

Try / catch

try:
    avg_loss = metric_logger.loss.global_avg
except AttributeError as e:
    avg_loss = None
    print(f"meter missing: {e}")

Prevention

When it happens

Trigger: Accessing e.g. metric_logger.losses or metric_logger.meterss, or reading a meter that was never set via __setattr__ (never assigned a SmoothedValue) before the first update call.

Common situations: Typos in training scripts, reading metrics before train_one_epoch populates them, divergence between meter names in train vs eval code paths.

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/92cfcf593cb491e7. Report an issue: GitHub.