# CIFAR100 image classification with Noisy Labels — Data

来源：https://www.kaggle.com/competitions/cifar100-image-classification-with-noisy-labels/data

采集时间：2026-10-19T11:28:38Z

## CIFAR100 dataset with noisy labels
Modified by Jonghyun Choi (2021.5.23)

## Note
1. Filenames are all randomized.
1. Some of the training samples' label are randomized for the noisy label
1. Testing samples' label are *not* randomized
1. Number of samples for training and testing are the same to the CIFAR100 dataset (except two files (one for training and one for testing) are removed from the original dataset for the duplicated label)
1. Now, labels of the testing samples are removed for submission.

### To use the dataset for the best test accuracy, plz do either
1. Split the training set into training and validation
2. Use cross validation