Automated Cleanup of the ImageNet Dataset by Model Consensus, Explainability and Confident Learning
This work addresses the problem of dataset quality for researchers and practitioners using ImageNet, though it is incremental as it builds on existing cleaning methods.
The paper tackled labeling mistakes and ambiguous images in the ImageNet dataset by developing automated heuristics based on model consensus, explainability, and confident learning, resulting in a 2-2.4% performance improvement for SqueezeNet and EfficientNet-B0 models.
The convolutional neural networks (CNNs) trained on ILSVRC12 ImageNet were the backbone of various applications as a generic classifier, a feature extractor or a base model for transfer learning. This paper describes automated heuristics based on model consensus, explainability and confident learning to correct labeling mistakes and remove ambiguous images from this dataset. After making these changes on the training and validation sets, the ImageNet-Clean improves the model performance by 2-2.4 % for SqueezeNet and EfficientNet-B0 models. The results support the importance of larger image corpora and semi-supervised learning, but the original datasets must be fixed to avoid transmitting their mistakes and biases to the student learner. Further contributions describe the training impacts of widescreen input resolutions in portrait and landscape orientations. The trained models and scripts are published on Github (https://github.com/kecsap/imagenet-clean) to clean up ImageNet and ImageNetV2 datasets for reproducible research.