Jayen Ashar   •   about 7 years ago

Misclassified training data?

Can we expect the test data not to be misclassified when scoring? I ask because I think there are a few misclassified samples in the training data. E.g., 6909.png is labelled as "Nectarines, peaches & apricots" but my classified and I think it should be "Berries & cherries".

Thanks,
Jay

• 1 comment
• 

  Willem Gerritzen Manager   •   about 7 years ago

  Hi Jayen,

  Thanks for your message. :) For now, a single image will correspond to a single class. In case there were 2 products in the image that show a defect, the label was chosen at random between the products. The number of images that concern this is minimal in the dataset and can also be considered as a noise factor.

  Best,

  Willem

Comments are closed.