RandAugment: Practical Automated Data Augmentation with a Reduced Search Space

Authors: Ekin Dogus Cubuk, Barret Zoph, Jon Shlens, Quoc Le

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Despite the simplifications, our method achieves state-of-the-art performance on CIFAR-10, SVHN, and Image Net. On Efficient Net-B7, we achieve 84.7% accuracy, a 1.0% increase over baseline augmentation and a 0.4% improvement over Auto Augment on the Image Net dataset. On object detection, the same method used for classification leads to 1.0-1.3% improvement over the baseline augmentation method on COCO.
Researcher Affiliation Industry Ekin D. Cubuk , Barret Zoph , Jonathon Shlens, Quoc V. Le Google Research, Brain Team {cubuk,barretzoph,shlens,qvl}@google.com
Pseudocode Yes Figure 3: Python code for Rand Augment based on Num Py.
Open Source Code Yes Code is available online.
Open Datasets Yes CIFAR-10 [14], SVHN [24], and Image Net [4] classification tasks.
Dataset Splits Yes N and M were selected based on the validation performance on 5K held out examples from the training set for 1 and 5 settings for N and M, respectively.
Hardware Specification No The paper mentions 'GPU hours' for Auto Augment's search, implying the use of GPUs, but does not provide specific models or other hardware details for their own experiments.
Software Dependencies No The paper mentions 'Num Py' and implies 'TensorFlow' through GitHub links, but no specific version numbers for any software dependencies are provided.
Experiment Setup Yes The Wide-Res Net models are all trained with K=14 data augmentations over a range of distortion magnitudes M parameterized on a uniform linear scale ranging from [0, 30]. Models are trained for 200 epochs on 45K training set examples. N and M were selected based on the validation performance on 5K held out examples from the training set for 1 and 5 settings for N and M, respectively.