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

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

Error message

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

What it means

AttributeError thrown by the SmoothedValue/MeterDictionary-style class's custom __getattr__. When attribute lookup fails on the instance, __getattr__ checks self.meters and self.__dict__; if the name is found in neither, it raises AttributeError naming the class and the missing attribute. This is the standard torch.utils.data distrbuted-utils pattern from maskrcnn-benchmark.

Source

Thrown at pytorch_segmentation/lraspp/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. Fix the attribute name typo to match a registered meter key
  2. Ensure the meter was created before access: any use of loss_meter.update(...) registers it via __setattr__
  3. Use getattr(obj, 'attr', default) or check `attr in obj.meters` before access

Example fix

// before
mean_loss = torch.stack([m.global_avg for m in metric_logger.meters.values()])
print(metric_logger.losss)
// after
print(metric_logger.loss)  # correct meter name registered via metric_logger.loss = SmoothedValue()
Defensive patterns

Strategy: validation

Validate before calling

if attr not in logger.meters and attr not in logger.__dict__:
    print(f"available meters: {list(logger.meters)}")

Type guard

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

Try / catch

try:
    value = logger.loss.global_avg
except AttributeError as e:
    value = float('nan')  # or fall back to default metric

Prevention

When it happens

Trigger: Accessing an attribute like logger.meters['loss'] misspelled (e.g. .los), or accessing a meter key that was never registered via __setattr__ on the MetricLogger before reading it, e.g. calling meter loss without a prior update().

Common situations: Typos in training scripts (meterr vs meters), reading a loss key before any training step registered it, or copy-pasting code between projects where the MetricLogger was populated with different meter names.

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/8d5befa845df134c. Report an issue: GitHub.