[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

Latest commit

 

History

History

noisy_labels

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 

Noisy Imagenette

This folder includes more information about the noisy version of Imagenette/Imagewoof, which is available with the current Imagenette/Imagewoof dataset as CSV files. Note that the Imagenette datasets with 5% and 50% noise are the main versions of the noisy dataset, with the leaderboards maintained here.

As shown in the Jupyter Notebook, the labels of the training set are randomly changed, while the validation set labels remain the same.

fastai's example training script provides arguments for noise levels (pct_noise) allowing you to train models with the desired noise level.

This folder contains the following files:

  • generate_labels.ipynb: The Jupyter Notebook used to generate the noisy labels (Noisy Imagenette CSV and Noisy Imagewoof CSV)
  • extended_lb.csv: This includes accuracy scores for all image sizes (128, 192, 256), number of epochs (5, 20, 80, 200), with an expanded set of noise levels (1%, 5%, 25%, 50%) on both Imagenette and Imagewoof, resulting in 96 scores. The CSV includes scores from this baseline.