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

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

Error message

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

What it means

This is the standard Python __getattr__ fallback on a SmoothedValue-style AverageMeter container: attribute lookups first check self.meters, then instance __dict__, and raise AttributeError naming the object type and missing attribute. Any code doing e.g. metric_logger.loss when 'loss' was never logged triggers it.

Source

Thrown at pytorch_keypoint/HRNet/train_utils/distributed_utils.py:137

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. Check the exact attribute name against the keys registered in meter_dict/update (print(metric_logger.meters.keys())).
  2. Fix the typo or log the missing metric before accessing it.
  3. Use getattr(metric_logger, name, default) for optional metrics.
  4. If this fires for dunder attributes during copy/pickle, add explicit __copy__/__deepcopy__/__getstate__ methods to the class.
  5. Return None instead of raising if optional semantics are desired (modify __getattr__ fallback).

Example fix

# before
print(metric_logger.losses)  # AttributeError: 'MetricLogger' object has no attribute 'losses'
# after
print(metric_logger.loss)
Defensive patterns

Strategy: type-guard

Validate before calling

available = set(metric_logger.meters.keys())
assert "loss" in available, f"'loss' never logged; available: {available}"

Type guard

def has_metric(logger, name: str) -> bool:
    return name in logger.meters or name in logger.__dict__

Try / catch

try:
    value = metric_logger.loss
except AttributeError as e:
    logging.warning("Metric missing: %s — using default", e)
    value = float("nan")

Prevention

When it happens

Trigger: Accessing an attribute on the metric logger that was never added via meters: metric_logger.some_metric before update_loss/meter_dict registered it; typo like metet_logger.lr when only 'lr' was tracked; accessing on a freshly constructed logger with no meters.

Common situations: Typos in metric names between train and eval loops; reading a loss key that differs across model versions; copying tutorial code whose logged keys don't match this repo's; __getattr__ interplay with copy/pickle probing __deepcopy__/__getstate__ which the container lacks.

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/f9c7226949460895. Report an issue: GitHub.