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

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

Error message

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

What it means

Identical family to error 150 but in the FCN project's distributed_utils.py: a MetricLogger/SmoothedValue container raises AttributeError when an accessed attribute is neither a registered meter nor in instance __dict__. The __getattr__ fallback is the last resort before failing, so the requested metric simply does not exist.

Source

Thrown at pytorch_segmentation/fcn/train_utils/distributed_utils.py:142

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. Compare the missing attribute name with keys passed to `metric_logger.update(...)` in train_one_epoch/evaluate
  2. Register the meter at logger creation (`self.meters[name] = SmoothedValue(...)`) or ensure update() is called before access
  3. Use `hasattr`/`getattr(..., None)` for optional metrics
  4. Check for typos and copy-paste drift between project variants

Example fix

// before
print(metric_logger.lr.global_avg)  # AttributeError: never registered
// after
metric_logger.update(lr=optimizer.param_groups[0]['lr'])
print(metric_logger.lr.global_avg)
Defensive patterns

Strategy: try-catch

Validate before calling

registered = set(metric_logger.meters) | set(metric_logger.__dict__)
assert 'loss' in registered, f"'loss' missing; registered: {registered}"

Type guard

def meter_exists(logger, name):
    return name in getattr(logger, 'meters', {}) or name in logger.__dict__

Try / catch

try:
    value = getattr(metric_logger, name).global_avg
except AttributeError as e:
    logging.warning("meter %s missing: %s", name, e)
    value = float('nan')

Prevention

When it happens

Trigger: Reading `metric_logger.<name>` for a key never passed to `update()`; mistyped meter name; accessing meters after `reset()` cleared them; code assuming FCN train loop registers meters that only the detection loop registers (e.g. 'mask_loss').

Common situations: Adapting the FCN train_one_epoch to log extra metrics without creating them; summary/print code copied from another project; running evaluation loop expecting train-only meters.

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/860999e6135d0060. Report an issue: GitHub.