Implementation:Kornia Kornia Accuracy Metric

Knowledge Sources	Kornia
Domains	Vision, Metrics
Last Updated	2026-02-09 15:00 GMT

Overview

Computes the top-k classification accuracy between predicted logits and ground truth targets.

Description

The accuracy function evaluates classification performance by computing the percentage of correct predictions among the top-k predictions. For each sample in a batch, the function selects the top-k classes with the highest logit values and checks whether the ground truth label is among them. The result is expressed as a percentage (0 to 100). This is a standard metric used in image classification tasks, particularly when evaluating models on benchmarks such as ImageNet where top-1 and top-5 accuracy are commonly reported.

Usage

Import this metric when you need to evaluate classification model accuracy, especially when computing top-k accuracy for multi-class classification problems. It accepts raw logits (unnormalized scores) and integer target labels.

Code Reference

Source Location

• Repository: Kornia
• File: kornia/metrics/accuracy.py
• Lines: 1-44

Signature

def accuracy(
    pred: torch.Tensor,
    target: torch.Tensor,
    topk: Tuple[int, ...] = (1,)
) -> List[torch.Tensor]:

Import

from kornia.metrics import accuracy

I/O Contract

Inputs

Name	Type	Required	Description
pred	torch.Tensor	Yes	The input tensor containing logits to evaluate. Shape is (B, C) where B is batch size and C is the number of classes.
target	torch.Tensor	Yes	The tensor containing ground truth class indices. Shape is (B,) or (B, 1).
topk	Tuple[int, ...]	No	A tuple of integers specifying which top-k accuracies to compute. Defaults to (1,).

Outputs

Name	Type	Description
result	List[torch.Tensor]	A list of scalar tensors, one per entry in topk, each representing the percentage accuracy (0-100) for that top-k value.

Usage Examples

import torch
from kornia.metrics import accuracy

# Single sample, top-1 accuracy
logits = torch.tensor([[0, 1, 0]])
target = torch.tensor([[1]])
result = accuracy(logits, target)
# result: [tensor(100.)]

# Batch with top-1 and top-5 accuracy
logits = torch.randn(32, 1000)  # 32 samples, 1000 classes
target = torch.randint(0, 1000, (32,))
top1, top5 = accuracy(logits, target, topk=(1, 5))
print(f"Top-1 accuracy: {top1.item():.1f}%")
print(f"Top-5 accuracy: {top5.item():.1f}%")

Related Pages

• Principle:Kornia_Kornia_Classification_Metrics